# WTF Is the Context Layer? The Missing Infrastructure for Production Agents

Prukalpa Sankar, Atlan

AI Engineer World's Fair 2026

[Talk page](https://ai.engineer/talks/8G_1-3IO4ZQ-wtf-is-context-layer-missing-infrastructure) · [Video](https://www.youtube.com/watch?v=8G_1-3IO4ZQ)

## WTF Is the Context Layer? The Missing Infrastructure for Production Agents

An agent can reason well and still use the wrong business definition. Atlan’s experiments show why shared context needs ownership, dependencies, review and a learning loop.

Familiarity with AI agents, business data systems and basic version-control concepts will help; no particular framework is required.

### Why doesn’t AI know your business?

[0:00](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=0s)

Why can an increasingly capable AI still struggle to do useful work inside a company? That is the problem behind Atlan co-founder Prukalpa Sankar’s context-layer question. Atlan works on giving AI the business knowledge it lacks, with customers Sankar names including GitLab, Zoom, Discord, Affirm, Mastercard and General Motors. She invokes Bill Gates’s formulation that content is king, then proposes its agentic-era counterpart: context will be king. By her account, 2026’s enthusiasm for context graphs reflects a problem that stronger models alone have not resolved.

In Sankar’s conversations, the question is often whether AGI is one year away or three. Inside a business, the question is whether greater capability produces better outcomes. Sankar illustrates the gap with a claimed progression from failing the bar exam to scoring in the top 1%. That chronology needs qualification: the [GPT-4 Technical Report](https://arxiv.org/abs/2303.08774) already reported **GPT-4** passing a simulated bar exam around the top 10% in March 2023; the talk does not identify the model or evaluation behind its top-1% figure.

Sankar says only one in five AI use cases reaches production, without identifying the underlying study. The displayed slide instead describes one in five reporting significant value, which is a different outcome. Her financial-benefit example has a firmer source: [PwC’s 2026 Global CEO Survey](https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-global-ceo-survey.html) found that 56% of 4,454 surveyed CEOs reported no significant financial benefit from AI through revenue or cost outcomes. That is narrower than saying AI delivered no benefit at all.

![Two panels compare 2023 and 2025 model benchmarks with claims of zero financial benefit for 56% of CEOs and one in five reporting significant value.](https://ai.engineer/api/talk-essay-media/a8f4f365ff88ee54e1eb0e827f67b5a909ea9a4aec47dfb78246a62933f654c6)

Model capability gains contrasted with limited business value from AI.

The human analogy is useful even without a precise equation. Sankar cites IQ as explaining only 10% of job-performance variance, without establishing the study behind that figure. But consider the practical distinction: the teammate with the highest SAT score is not necessarily the one who works most effectively, incorporates feedback or learns fastest. **Real-world performance depends on both intelligence and context**—the knowledge, skills and expertise acquired on the job.

Sankar describes intelligence as having grown 1,000× over a decade and 2× in six months, without specifying a measurement scale. Her architectural point is that business context has not advanced in parallel. Moving data to the cloud does not extract the situated knowledge still trapped in dashboards, Slack threads or the head of an analyst who might leave next week. To understand what agents need, start with how that analyst learns to work.

### A simple metric question requires three kinds of context

[4:25](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=265s)

Maya is the fictional data analyst at McContext Burgers whom everyone messages when they need an answer. One morning, a franchisee asks why drive-through time is up this week. The question sounds like a straightforward comparison. Before querying anything, however, Maya has to establish what the comparison means.

| Context | What Maya needs |
|---|---|
| Knowledge | Metric definition, requester, weekly cutoff and time zone |
| Expertise | A diagnostic sequence for plausible causes |
| Norms | The scope and form of an appropriate answer |

First, finance and operations may mean different things by drive-through time. The week might run Monday through Sunday, but even that leaves Pacific versus Eastern Time unresolved. These are business facts: the map that makes a query meaningful. Next comes the diagnostic playbook. Maya knows to check third-quarter weather seasonality and the product launched in the previous quarter. Finally, she scopes the response to the person asking. Her useful output includes the explanation and root cause, not just a changed number.

Maya joined only a year earlier. Initial training gave her a start; shadowing an experienced teammate taught her why particular checks mattered. Mistakes, manager feedback and unfamiliar edge cases refined that judgment. Building an agentic Maya therefore requires more than packaging an onboarding document. It requires a way to acquire, correct and retain expertise through work.

### Specialist agents worked until their context diverged

[7:23](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=443s)

Atlan’s first approach, roughly 18 months before the talk, began with a jobs-to-be-done map of its customer-experience team. Which activities filled the day, and how much could AI help with each? Documentation and meeting preparation looked promising. Relationship management looked much less amenable to automation. The team used those judgments to assign scaling factors and build narrowly focused agents. Hermione became the health intelligence lead; Moneypenny became the financial risk analyst. Each agent was optimized to do one thing well, and initially that worked.

![A Customer Success Manager sits above connected agent cards for call intelligence, health intelligence, onboarding, and specialist analyst roles.](https://ai.engineer/api/talk-essay-media/194fa85549a3d4d66ac50b7ee2ef5030cf2d6f46a36b40970e65ea3c7aec3960)

“One human, ten specialized agents”: a customer-success agent hierarchy.

By the middle of the preceding year, Sankar says building an agent took about five minutes, while preparing the business context needed for accuracy remained slow. Agent quality depended heavily on context-engineering quality, and mistakes cost stakeholder trust. The difficulty shifted from creating an agent to keeping it correctly informed.

A positioning change exposed the coordination problem. In a human organization, marketing can announce new positioning at a town hall, and sales development representatives know to change their pitch. Atlan’s marketing agents received changes, but its website SDR agent kept pitching the old version. The agents had no equivalent shared communication infrastructure, and the team lacked visibility into how their contexts were connected. When something went wrong, it was hard to distinguish a model problem from an agent problem or a context problem.

Separate memory systems made the problem worse. Agents learned independently and differently, making a single version of truth difficult to establish. Portability then added another failure mode. Over the preceding year, the team moved from Relevance to Google ADK, then tried Glean, moved to Claude Code at the start of the year, and eventually used roughly equal parts **Claude** and **Codex**. Each transition left context trapped in the previous system. The emergence of general-purpose agents prompted a different question: could the business knowledge live independently of the agent running the task?

### Separate shared expertise from the agents that use it

[11:15](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=675s)

Maya’s effectiveness also comes from her team. Strong teams share language, a picture of what is true today, playbooks and decision rights. They learn together. A failed launch becomes collective memory rather than a lesson each person must rediscover. Atlan’s revised approach translated that arrangement into domain experts responsible for skills, a common company brain holding their contributions, and retrieval mechanisms that make those contributions available to general-purpose agents. This shared infrastructure is the **context layer**.

The marketing experiment connected data systems, social and community platforms, advertising platforms and analytics. Its agent side deliberately remained open: Claude Code and Cowork, an internally deployed Claude interacting in Slack, and external products including Qualified and Artisan. Between those systems and agents sat the shared repository. The best SEO practitioner contributed an SEO skill; the competitive-intelligence expert contributed that domain’s skill. Expertise could be maintained in one place and used by different agents.

The repository needed more than instructions:

- **Data graph:** An autonomous ads agent doing daily analysis needs to know which tables to query.
- **Skill library:** Domain playbooks describe how to perform the work.
- **Semantics and metrics:** Definitions establish how ARR is measured and what counts as a qualified lead at this company.
- **Organization and entities:** Organizational structure and business entities supply additional context.

Sankar reports that the marketing team built about 300 skills and 40 agents over six months. Those counts describe the experiment’s scale; they do not measure its accuracy.

### Shared skills need dependency and quality management

[14:29](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=869s)

Centralizing context did not eliminate operational problems. It made their structure clearer. A competitive-intelligence skill learns what is changing in the market. Its output feeds category positioning, which feeds the sales battle card. Each skill can improve locally while disrupting a downstream consumer. **Context needs to be managed like code because changes have dependencies.**

A minimal JSON representation makes the dependency direction explicit: each skill names what it consumes. This illustrates the chain described in the talk, rather than an Atlan API.

```json
{
  "skills": [
    {
      "id": "competitive-intelligence",
      "dependsOn": []
    },
    {
      "id": "category-positioning",
      "dependsOn": ["competitive-intelligence"]
    },
    {
      "id": "sales-battle-card",
      "dependsOn": ["category-positioning"]
    }
  ]
}
```

A change to `competitive-intelligence` has a direct consumer in `category-positioning` and an indirect consumer in `sales-battle-card`. Recording those relationships gives a reviewer a concrete impact path to inspect.

The team also encountered outdated skills, drift and unclear ownership of quality. Security and governance became difficult: Sankar describes secrets hardcoded in `.env` files and people downloading public skill repositories. Portability across agent systems remained necessary. A common repository solves distribution only partially; it still needs responsibility for what enters it, what changes and who can rely on it.

![Four cards list skill dependency management, skill quality ownership, security and governance, and context portability.](https://ai.engineer/api/talk-essay-media/be4725a964a25b30c419ef0d207e01ffa2e6c81883d768416d66e958b17eb60c)

Four challenges in managing context like code.

### Give context a review flow—and learn from traces

[15:31](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=931s)

Sankar frames the next infrastructure question as what a GitHub for context would look like. Company knowledge needs lifecycle management, collaboration and versioning. That includes deciding which context is local, which is global and how each stays current. A proposed skill PR flow would combine learning with quality checks and security posture management, rather than treating every edit as immediately trustworthy.

Dependency visibility should answer what else a change affects. Explicit roles should identify the approver, maintainer and contributors. Together, these capabilities would create a human-plus-AI workspace in which shared skills can evolve without losing accountability. These are the capabilities Sankar proposes for the context layer, not a claim that every part was already implemented.

Agent interactions then become inputs to that development process. Sankar describes using a specialized harness to read traces and reconstruct candidate improvements. The important boundary is between discovering a possible lesson and accepting it into shared context:

1. Read the traces of agent work.
2. Reconstruct candidate context or skill changes.
3. Present those changes to a maintainer.
4. Have the maintainer approve or reject them.

The learning loop compounds through reviewed improvements. An observed interaction supplies evidence for a change; it does not automatically authorize that change.

### Reconstruct the context already inside business systems

[17:13](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1033s)

Where should a company begin if its business is spread across dozens of systems? Sankar poses the example of an organization with 60 systems, then points to context already embedded in their connections. Salesforce, HubSpot, the data warehouse and the application layer each contain parts of the business’s operating logic. Context gets lost at the hops between them. Reconstructing those relationships gives AI a basis for seeding the first company brain.

Sankar reports strong accuracy from this reconstruction approach, but supplies no numerical evaluation in the spoken explanation. The slide connects an Enterprise Data Graph and an Enterprise Semantic Graph to column descriptors, joins, filters, business questions, metrics, KPIs, entities, relationships and a bootstrapped ontology. Its visible 89% callout has no established metric or evaluation conditions in the supplied explanation, so it cannot serve as an accuracy result here. The mechanism is the useful part: recover relationships across systems before asking an agent to reason over isolated fragments.

![Enterprise Data Graph and Enterprise Semantic Graph boxes connect to column descriptors, joins, filters, business questions, metrics, KPIs, entities, relationships, and a bootstrapped ontology. A left panel highlights 89%.](https://ai.engineer/api/talk-essay-media/adad2f2024d0f7b6dd3e782aeb751bbda2b8b0b25416f98969d776654bda5dc5)

Bootstrapping context from enterprise data and semantic graphs.

### The context layer is a continuing loop—and company IP

[18:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1085s)

The resulting architecture turns Maya’s knowledge, expertise and norms into machine-usable context. It continually mines business systems and feeds a shared company brain. Teams develop and maintain skills through a context lifecycle as they deploy agents. Agents retrieve what they need through MCP, SQL, vector retrieval or hybrid assembly; traces return evidence to the learning loop. Retrieval is one part of the architecture, alongside acquisition, maintenance and feedback.

Hardcoding business context into individual agents makes that loop difficult to sustain. The familiar disagreement in which sales and finance report different revenue numbers becomes more consequential when autonomous systems act on those definitions. Without shared context and visible ownership, organizations risk reproducing their inconsistencies at machine speed and scale.

**Context is also intellectual property.** If competitors can use the same models, their distinctive capabilities must come partly from how they conduct business. A customer-support agent at American Express should not be interchangeable with one at Amazon: each should embody its company’s practices, culture and norms. Sankar closes by inviting collaboration with teams building company brains. The infrastructure challenge is to make that accumulated expertise usable by autonomous systems while preserving the judgment that makes it the company’s own.

## Resources

- [The Context Layer: Knowledge Graph's second act](https://atlan.com/context-and-chaos/issue/the-context-layer-knowledge-graphs-second-act/) — Sankar’s companion argument about organizational knowledge, skills and the context layer, developed for the Knowledge Graph Conference.
- [Building a Company Brain for AI Agents with Shared Context](https://www.linkedin.com/posts/prukalpa_by-middle-of-last-year-building-an-agent-activity-7483883266668732416-gQtA) — The speaker’s recap of Atlan’s shared-context marketing experiment and the framework changes that motivated it.
- [PwC’s 2026 Global CEO Survey](https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-global-ceo-survey.html) — Survey findings on CEOs’ reported AI revenue and cost outcomes, including the 56% reporting no significant financial benefit.
- [GPT-4 Technical Report](https://arxiv.org/abs/2303.08774) — OpenAI’s original report includes a simulated bar-exam result around the top 10% of test takers.
- [Build a context layer with Atlan](https://docs.atlan.com/agents/how-tos/build-a-context-layer) — Current setup guidance for a scoped context layer using Atlan connectors, expert review and MCP; requires an Atlan workspace.
- [Building ADK Agents with Skills](https://developers.googleblog.com/en/developers-guide-to-building-adk-agents-with-skills/) — Google’s practical guide to loading reusable expertise through progressive disclosure and several skill packaging patterns.

## Slides

### WTF is the Context Layer

The talk introduces the context layer as missing infrastructure for production agents.

[0:39](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=39s) [Atlan title slide: The Missing Infrastructure for Production Agents.](https://assets.ai.engineer/sites/aiecode2025/sha256/00d3e1ccc10a9d29089d72b5b5e526533df20c4fe501887a6a00f0fb4cb5bad9.webp)

atlan

WTF is the Context Layer

The Missing Infrastructure
for Production Agents

### AI doesn’t know your business, we fix that

Atlan says it helps AI understand a business and works with companies ranging from GitLab, Zoom and Discord to Mastercard and General Motors.

[0:55](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=55s) [Atlan slide displaying a grid of customer logos.](https://assets.ai.engineer/sites/aiecode2025/sha256/c517cc7a267f10c60a1cd9844a5c0e5bd50587c0a157f976cd09ffa007d2321d.webp)

AI doesn’t know your business, we fix that

atlan

mastercard
GitLab
zoom
HubSpot
elastic
Nasdaq
RIOT GAMES
Medtronic
GRAINGER
FOX
affirm
Virgin media O₂
workday
Discord
gm general motors
Dropbox

### Content is King, Context is King

The speaker recalls presenting “context will be king” for the agentic era, alongside Bill Gates’s 1996 “Content is King” message from the internet era.

[1:16](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=76s) [A Bill Gates image labeled Content is King beside a photograph of a speaker in front of Context is King lettering, dated June 2025.](https://assets.ai.engineer/sites/aiecode2025/sha256/d55c84fa00d868239b56fc22d4e55a3ad3ffca1c98ebd21c61f2301108dd69b0.webp)

Content is King
Bill Gates 1996

Context is King.

Jun 2025

atlan

### 2026 is the year of Context

A collage shows context appearing in AI operating systems, enterprise data, graphs and engineering roles. The speaker says new examples keep appearing every few days.

[1:39](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=99s) [A collage of posts and headlines about context layers, context graphs, and AI agents.](https://assets.ai.engineer/sites/aiecode2025/sha256/2a9263346c09473adb66e860834a924cbc7926375c5d2ac49a8f5de21c1a24bb.webp)

2026 is the year of Context

Y Combinator
@ycombinator

The AI Operating System for Companies
@sdianahu

The best AI-native companies have made their entire company queryable: every meeting, ticket, and customer interaction legible to an intelligence layer that learns from it.

Building this today requires brutal integration work, and there's no product that connects all this context into a single layer that can reason across it.

[illegible] Raises USD 3 Million to Build Context Layer Enterprise AI

[illegible] Launches Context Mesh to Connect Enterprise Data to AI Agents

Context Graphs: Capturing The "Why" In The Age of AI
Dharmesh Shah
Founder and CTO at HubSpot. Helping millions grow better.
January 10, 2026

Jaya Gupta @JayaGup10 · Dec 23, 2025
X Article
AI’s trillion-dollar opportunity: Context graphs
The last generation of enterprise software created a trillion-dollar ecosystem by becoming systems of record. Salesforce for customers. Workday for employees. SAP for operations. Own the canonical...

[illegible] to Deploy 1,000 Context Engineers, Powered by [illegible], to Industrialize Agentic AI

Interface text:
396 · 1.3K · 7K · 4.8M

### Models are getting smarter, but not more useful

The slide contrasts rising model scores from 2023 to 2025 with reported business value: ~56% of CEOs cite zero financial benefit from AI, while Gartner is cited for 1/5 reporting significant value.

[1:48](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=108s) [A table compares 2023 and 2025 model scores alongside reported measures of financial benefit and value from AI.](https://assets.ai.engineer/sites/aiecode2025/sha256/6e8f462bd453806d201ffb792ee8b52764d1bf8179cd072b3fdef17d279c78a1.webp)

Models are getting exponentially smarter

2023 | 2025
Graduate science | 35% | 87% (experts: 65%)
Competition math | 9% | 97%
Novel reasoning | 5% | 87.5% (humans: 85%)
Software engineering | 15% | 60%+

But not exponentially more useful

0 Financial benefit from AI
~56% of CEOs

1/5 Report significant value
~Gartner

atlan

### Cognitive intelligence and real-world effectiveness

The scatter plot shows a slight upward trend with wide variation; the slide attributes ~10% of performance variance to IQ.

[2:39](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=159s) [A scatter plot illustrates the slide’s claim that IQ explains about 10% of performance variance.](https://assets.ai.engineer/sites/aiecode2025/sha256/ed87387494aee50115f4960d24bf621442f563f57b892586585769619b216eb4.webp)

Cognitive Intelligence (IQ) doesn’t determine real world effectiveness.

Y-axis: Real world effectiveness
Y-axis labels: Low, High
X-axis: Cognitive intelligence (IQ)
X-axis labels: Low, High

~10% of performance variance explained by IQ

Based on: Sternberg & Wagner (1986), Practical Intelligence; Sackett, Zhang, Berry & Lievens (2022), Journal of Applied Psychology.

atlan

### Performance as a function of intelligence and context

The slide models performance as a function of intelligence and context: cognitive horsepower combines with the knowledge, skills, and tools built through doing the work.

[3:09](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=189s) [The formula P = f(I, C), with definitions of performance, intelligence, and context.](https://assets.ai.engineer/sites/aiecode2025/sha256/3d255d8158a0f98d1c745effdfcbbb98234b75179295bde9bcd479e114a38be2.webp)

P = f(I, C)

Performance: The outcomes you actually deliver in the real world.

Intelligence: Cognitive horsepower. How fast you learn, reason, process. Largely fixed.

Context: knowledge, skills, and tools built through doing the work.

atlan

### The last decade

The speaker contrasts a roughly 1,000x rise in intelligence over the last decade with business context that barely moved, still locked in dashboards, Slack threads, and analysts’ heads.

[3:34](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=214s) [A rising intelligence curve contrasts with a flat context line; the slide describes intelligence as compounding and business context as barely moving.](https://assets.ai.engineer/sites/aiecode2025/sha256/e368a468a55a7c225889c77762335e612a25834ed09aa867b25fe0a5f0ae1165.webp)

The last decade

Intelligence
~1000x Compounding. Commoditizing.

Context barely moved.

Intelligence (I) — frontier capability. Compounding. Commoditizing.

Context (C) — Situated knowledge of your business. Locked in dashboards, Slack threads, and the analyst who’s about to leave.

atlan

### Helping AI build business context

To help AI build business context, the speaker looks to how people are trained and onboarded.

[4:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=245s) [The slide asks how to help AI build context about a business and suggests training and onboarding it as humans are trained.](https://assets.ai.engineer/sites/aiecode2025/sha256/45908fb85f408c86052a86563c26d8bc1d20dc02fd6d4133df5503102fb5fd0b.webp)

How do we help AI build context about our business?

Hint: The same way we train and onboard humans.

atlan

### Meet Maya

Maya, a data analyst at McContext Burgers, gets a drive-through question from Marcus, a franchisee with 14 stores. Answering it requires context she carries in her head.

[4:25](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=265s) [Illustrated analyst Maya receives a franchise owner's question about an increase in drive-through time.](https://assets.ai.engineer/sites/aiecode2025/sha256/be3b530c01adfd0a06a39b54de911fad01625879464868092c2ad0773494318e.webp)

Meet Maya

She's a data analyst at McContext Burgers — the person every operator pings when something looks off. This morning it's Marcus, a franchisee with 14 stores

Marcus
Franchise owner

Why is my drive-through time up this week?

Sounds simple. It isn't.
Before Maya can even start, she has to answer a stack of questions the message doesn't — and none of them are in the data. They're in her head.

atlan

### Knowledge, expertise, and norms

Answering the drive-through question combines definitions of the metric and week, a diagnostic playbook, and rules about which data each role can see and use.

[5:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=305s) [A Venn diagram describes knowledge, expertise, and norms as overlapping types of context needed to answer a drive-through question.](https://assets.ai.engineer/sites/aiecode2025/sha256/997c6cacd5ceca2f95484f89565c60ce9a1f7f5e085e6df96b53721fbb256728.webp)

Why is drive-through time up this week?

To answer this seemingly simple question, the agent needs multiple types of context

KNOWLEDGE
The map of the business

EXPERTISE
How work gets done

CONTEXT

NORMS
The rules of acceptable action

"Drive-through time" means avg_dt_secs, not the POS number Finance uses.

"This week" = Mon–Sun in store local time, closed Sunday 11:59 PM. Not a rolling 7-day window.

Diagnostic playbook — validate the premise, check seasonality, then investigate root causes.

Root-cause correlation — cross-reference against product launches, weather, public events.

Persona scoping — Store manager only sees his store data. VP Ops sees the full chain.

Metric authority — Store manager gets avg_dt_secs; Finance gets POS metrics.

KNOWLEDGE
The map of the business
What things mean. Entities, definitions, metrics, relationships — the ontology of how the company talks to itself.

EXPERTISE
How work gets done
The procedural know-how. Playbooks, workflows, task decomposition — the practical skill of running the business.

NORMS
The rules of acceptable action
What is allowed. Policies, permissions, scoping, approval paths — the guardrails that keep agents safe.

atlan

### Maya nails it

Maya traces a roughly 18-second rise in drive-through time at six single-lane stores to the Double Stack’s extra prep step. She recommends pre-staging patties during the 11–1 rush; dual-lane locations are flat.

[6:14](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=374s) [Maya reports that drive-through time increased at single-lane stores, attributes it to a product launch, and suggests pre-staging patties.](https://assets.ai.engineer/sites/aiecode2025/sha256/c94abf1587217d547e575f174c7d3a5e39f8a870cf6f2276777175a3659e4525.webp)

Maya nails it

Scoped to his stores. Root cause found. Fix attached.

Maya
Looked into it

Your drive-through time is up ~18s, but only at your 6 single-lane stores — dual-lane locations are flat.

Cause: the Double Stack launched Monday. Extra prep step, no second staging lane to absorb it.

Fix: pre-stage Double Stack patties at single-lane stores during the 11–1 rush. Should claw most of it back this week.

Marcus
Franchise owner

This is exactly what I needed.

atlan

### How Maya learned to be a great analyst

Maya learns the business vocabulary and tools, then develops judgment by shadowing a teammate, making a mistake, receiving feedback, and mapping an edge case the wiki missed.

[6:26](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=386s) [Comic panels trace Maya’s learning through vocabulary, tools, shadowing, mistakes, feedback, and an edge case.](https://assets.ai.engineer/sites/aiecode2025/sha256/0b7b21afea2bee7a0bd8f60f5d05f395c8756d00c662e92549522d5292265bbb.webp)

How Maya learned to be a great analyst

Week 1
‘Drive-through time’ = avg_dt_secs. Not the Finance number.
The vocabulary
What's true: the metrics, the definitions, the map of the business.

Week 2
Clicked through every dashboard. Sample queries ran clean.
The tools
System fluency: where the data lives and how to pull it.

Week 4
Prove it’s real. Rule out seasonality. Then go hunting.
Shadowing
The diagnostic playbook: validate → check seasonality → root-cause.

Month 1
I sent the what. Marcus needed the why. Never again.
the what.
Double Stack launch.
First wrong answer
The answer is the why, not the what — the Double Stack jams single-lane stores.

Month 2
You showed him chain-wide. He’s a franchisee — 14 stores only.
Manager Feedback
Persona scoping & metric authority: Ops gets avg_dt_secs; Finance gets POS.

Month 3
Docs had three lines. So I built my own map.
Edge Case
Skills- Knowledge (The cases and people that wiki doesn’t cover)

### How can we build the agentic Maya?

Atlan introduces its experiments and learnings from building an agentic Maya.

[7:11](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=431s) [Section slide introducing experiments and learnings about building an agentic Maya.](https://assets.ai.engineer/sites/aiecode2025/sha256/f16eea9fc320a7b8b664cff124f1dad32b454421c668d67081b725639a815cc9.webp)

So how can we build the agentic Maya?
Our experiments & learnings

atlan

### Era 1: Bootstrapping Agents

Atlan began bootstrapping agents roughly 18 months before the talk, starting with a jobs-to-be-done analysis of its customer experience team.

[7:22](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=442s) [Section slide titled Era 1: Bootstrapping Agents.](https://assets.ai.engineer/sites/aiecode2025/sha256/f8d163a122589bd5fc94a7170eb79a0276e3f3b2dc9d1a4b8575969a5d71ee61.webp)

Era 1:
Bootstrapping Agents

atlan

### Jobs to be done, split between humans and AI

Atlan mapped customer experience tasks by hypothesized AI involvement: documentation and meeting prep lean AI-led, while relationship management stays human-led. The map also identifies two new AI-related tasks.

[7:33](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=453s) [A chart places customer experience tasks along a human-led to AI-led spectrum, with two new agent-management tasks.](https://assets.ai.engineer/sites/aiecode2025/sha256/7d38c744a3e06a4259d06f409b96313b81af984766f9caac1c5fe8093e240cf8.webp)

Example: customer experience team

Jobs to be done, split between humans and AI

HUMAN-LED                                  AI-LED

Documentation & Meeting Prep
Risk Detection
Technical Advisory & Solutioning
Strategic Decisioning
Relationship Management
Signal Identification & Discovery
Write Scripts & Building Apps
Managing AI agents — NEW
Gathering & Actioning Signals from AI — NEW

atlan

### One human, ten specialized agents

A customer success manager oversees specialized agent leads and analysts. The agents focus on individual topics, from customer health and financial risk to product usage and support.

[8:13](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=493s) [An organizational chart connects a customer success manager with specialized agent roles.](https://assets.ai.engineer/sites/aiecode2025/sha256/2a89ec8e1db907800540712ada52ff9548a00a0af6923b7572468cc49bad7db3.webp)

One human, ten specialized agents

HUMAN
Customer Success Manager
Manages all the leads

AGENT
Donna, Call Intelligence Lead
Manages a team to ensure all the tasks after a call are performed well.

AGENT
Hermione, Health Intelligence Lead
Reviews current state, identifies risk and opportunities, and recommends next steps.

AGENT
Sheldon, Customer Onboarding Lead
Manages a team to ensure onboarding to C1 stage is performed at the highest quality.

AGENT
Sherlock, Strategic Success Analyst
Success plan management, value articulation, milestone tracking.

AGENT
Leslie, Relationship Analyst
Stakeholder engagement, champion analysis, communication patterns.

AGENT
Json, Product Analyst
Deep dive on a customer's product usage data to find insights and patterns.

AGENT
Moneypenny, Financial Risk Analyst
Deep dive on a customer's commercial health from contracts and external financial signals.

AGENT
Nora, Support Analyst
Support ticket trends and current status of open tickets.

AGENT
Linkin, Relationship Analyst
Gathers LinkedIn profile details and recent posts based on a person's name and company.

atlan

### Challenges with this approach

As the agent fleet grows, context becomes harder to prepare, govern and trace. An SDR agent kept pitching old positioning, while separate learning and platform changes left context fragmented or trapped.

[8:40](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=520s) [Three columns describe context engineering, context lifecycle and governance, and context sprawl and drift.](https://assets.ai.engineer/sites/aiecode2025/sha256/ddb74012c9acdb105d82c7d36ac23d1ec718474c5e25634bcaca9e5ed1becf60.webp)

Challenges with this approach

01
1–3 AGENTS
Context Engineering

How fast can you get to a trusted V1 that is accurate?

“Building the agent takes 5 minutes. Giving it business context to make it accurate takes forever

Quality of agent depends on quality of context engineering - leading to tons of “lost trust” cases with business stakeholders.

02
3–30 AGENTS
Context Lifecycle Mgmt. & Governance

How do I collaborate with business experts on context? Who owns, approves, maintains it?

“Marketing changed positioning last quarter. The SDR agent still pitches the old version. Nobody knows what else depends on it.”

“When an agent gets it wrong, I can’t trace which piece of context caused it — let alone roll it back.”

03
30+ AGENTS
Context Sprawl & Drift

How do I keep one truth — portable — across platforms that each learn their own way?

“When one agent learns, nothing passes back to a shared layer. Now every agent speaks a slightly different version of the truth”

Context Lockin We have gone through Relevance → Google ADK → Glean → Claude Code → Claude + Codex. Context gets trapped in each iteration.

atlan

### A different approach with general-purpose agents

After context became trapped in successive systems, the team began considering a different approach using general-purpose agents.

[11:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=665s) [Section slide asks whether general-purpose agents offer a different approach.](https://assets.ai.engineer/sites/aiecode2025/sha256/c009c4decbd2a53caef771d50b345f8f89481aa90d9ecf4060b4009b1f5c48ee.webp)

Era 2:
What if there was a different approach with general purpose agents?

atlan

### Maya is part of a team

Maya works alongside Diego in ads, Priya in store operations, Sam in customer support, and Lena in supply chain. Each brings a different view of the business.

[11:14](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=674s) [Illustrated team with descriptions of Diego, Priya, Sam and Lena’s roles.](https://assets.ai.engineer/sites/aiecode2025/sha256/97d0fea0f2fdf7de2e70edb5f1796472e7b18c3313fe6bf985523e92ab6ccccc.webp)

Maya isn’t an individual star, she’s part of a team

Priya
owns store operations. She knows which stores are single- vs. dual-lane.

Sam
runs customer support. He's hearing the wait-time complaints.

Diego
runs ads. He launched the Double Stack promo.

Lena
manages supply chain. She knows what's in every kitchen.

atlan

### Dream teams are built on shared context

Shared knowledge, playbooks and decision norms rest on a common context layer. Shared judgement and memory form a learning loop, with lessons becoming part of the playbook.

[11:29](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=689s) [Six cards describe shared language, current understanding, playbooks, norms, judgement and memory above a shared context foundation.](https://assets.ai.engineer/sites/aiecode2025/sha256/afe7bc841a6f23215c5dce73b3080b5df9a407747a0c3e45e2068c66bcef5a05.webp)

Dream Teams are built on shared context

Knowledge

01
Shared language
Everyone means the same thing by “drive-thru time,” “active customer,” “a good week.”

02
A shared picture of what’s true today
The Double Stack just launched and Q2 is the biggest season — expect higher traffic.

Expertise

03
Shared playbooks
How work gets done — Maya checks with Sam & informs Priya about the drive-through time spike.

Norms

04
Shared norms
Who’s allowed to make which decision — only Priya makes the final decision on drive-through.

Compounding Learning Loop

05
Shared judgement
What “good” looks like and what we’d never do — Sam knows what he can offer a customer without asking.

06
Shared memory that compounds
Learnings become part of a playbook — rain is the first thing to check when drive-through time spike.

THE FOUNDATION
A shared layer of context

atlan

### A company brain for a general-purpose agent

The proposed model brings team skills into a shared company brain, then connects that context layer through retrieval to a general-purpose agent.

[12:01](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=721s) [HR, finance and marketing skills feed a company context layer, which connects through retrieval to a general-purpose agent.](https://assets.ai.engineer/sites/aiecode2025/sha256/38a9077356e2900251af1b7db61e290df227d32acb521ff448e7a02a750858f9.webp)

What if there was a different way?

TEAMS
HR
People & policy

Finance
Books & controls

Marketing
Brand & demand

DOMAIN SKILLS
HR: Skill 1, Skill 2, Skill 3
Finance: Skill 1, Skill 2, Skill 3
Marketing: Skill 1, Skill 2, Skill 3

ONE PLACE, ONE AGENT
Company Brain
Context Layer

Retrieval

General-purpose agent
One agent that answers any question, backed by every team's skills.

atlan

### Marketing context layer architecture

Atlan’s marketing experiment connects its systems and multiple agents to a shared context layer of skills and knowledge. Teams contribute expertise to a common repository that agents can use across marketing work.

[12:40](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=760s) [Diagram connects marketing systems to agents, a shared layer of skills and knowledge, supporting infrastructure and outputs.](https://assets.ai.engineer/sites/aiecode2025/sha256/979b6bbcbf57185b48593273c48782dad3237260d511d392fa39fc1fbf91a6e7.webp)

Internal systems
CRM, Calls, Knowledge
Hubspot
Salesforce
Gong
Glean
Demandbase
Granola
Slack
Common room

Analytics
Traffic, Rankings, Search
Google Analytics
Search Console
MS Clarity
Serp API
PageSpeed

Ad Platforms
Campaign data
Google Ads
Linkedin Ads
Reddit Ads

Social & Community
Conversations, Signals
Reddit
Linkedin
X/Twitter
Newsletters
Web Search

Data
Warehouse, Documents
Snowflake
Google Sheets
Site maps

Context Layer
Shared, version controlled, compounding. Populated by humans and agents, together.

Skills
How to do things—automic, reusable, compounding.
SEO & AEO
Content
Analytics
Campaigns
Competitive intel
Creative
Enablement
PMM
Event
Outreach
Nurture
Reporting
Social
Website
Copy & voice
Lead signals

Knowledge
What to know and why now—cross pollinates across pods.
Brand voice
Design system
Playbooks
Competitors
Market intel
Customer stories
Pricing
Category narrative
Product narrative
Msg pillars
Sales plays
Style guide
Glossary
ICP

Agents
Any agents plugs in. Same context, same skill. Different trigger, different runtime.

Claude
Code, co-work
Interactive, human triggers

Atlan Claw
Market research, CWV fixes
Autonomous

Qualified/ artisan
AI SDR
Real time, visitor triggers

Model Gateway
Multi-model routing

Memory
Persists across sessions

Evals
Quality gates before shipping

Secrets
Auto-injected at runtime

Content
SEO, Thought Leadership, Case Studies

Campaigns
Ad, Email, Social—Coordinated

Enablement
Battlecards, Positioning, Demo scripts.

Reports
Dashboard, Analysis, Intelligence.

Creative
Image, Video, Design assets.

Website
Pages shipped live via atlan.com

atlan

### Impact in the last 6 months

In six months, the team created 300+ skills and 40+ agents. The examples range from research and brand-building skills to agents for product adoption, content, events, website optimization and campaigns.

[14:14](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=854s) [Two lists highlight more than 300 skills and 40 agents created, with examples of each.](https://assets.ai.engineer/sites/aiecode2025/sha256/2e28c274d032404615c9880660f3e463527ec80cc45cd41ebbe7c9d0af3f3186.webp)

Impact
in the last 6 months

300+ Skills created
/deep-research
Research topics across sources & synthesize findings
/design-council
Run the full double-diamond design process
/report-writing
Structure, write and publish reports from raw data
/brand-kit
Build & maintain a single source of brand truth
/community-architect
Design & build community programs from the ground up

40+ Agents created
Atlas
Drive & analyse product adoption
Helix
Research, write, and ship content at scale
Accio
Plan, run, and measure events end-to-end
Optimus
Optimize every pixel & maintain a high performance website
Focal
Build and manage campaigns across digital platforms

atlan

### Context needs to be managed like code

As skills evolve, changes in competitive intelligence can break positioning and sales battlecards downstream. The team also faced unclear quality ownership, exposed secrets, and difficult context portability.

[14:26](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=866s) [Four cards describe skill dependencies, quality ownership, security and governance, and context portability.](https://assets.ai.engineer/sites/aiecode2025/sha256/39abd1507a626717d19a85737130c603a457690efb69bc74cd7ff174b18eb9a5.webp)

THE PROBLEMS WE FOUND

Context needs to be managed like code

01
Skill dependency management
Competitive Intel skill → Category positioning skill → sales battlecard skill; each skill is learning and evolving, but breaks downstream skills. Skills get outdated and drift.

02
Who owns skill quality?
No one has a clear view of what exists, who owns it, who improves context engineering, what is safe to use, or what is actually running.

03
Security & governance nightmare
Secrets hardcoded in local .env files across 60+ systems. Agents see raw keys. People download public skill repos without security management.

04
Context Portability
Claude Code, Codex, Artisan/ Qualified, Atlan Claw - portability of context across humans, desktop use, and multi agent systems was complicated

atlan

## Full transcript

[0:00](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=0s) [upbeat music] Hi everyone, uh, my name is Prukalpa. I'm the founder of Atlan.

[0:17](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=17s) Um, and, uh, today I'm gonna talk about this thing where context is having its moment. Uh, and so my goal today is to talk about, like, WTF is the context layer?

[0:29](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=29s) Um, just before I start, and I promise this is the last time...

[0:35](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=35s) Um, I don't know if the clicker is working.

[0:57](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=57s) Atlan, we... It's, it's working. Yeah. Thank you. Um, the problem we solve is we say AI doesn't know your business. We've fixed that. We work with an incredible group of companies around the world, ranging from GitLab and Zoom and Discord and Affirm to large enterprises like Mastercard and General Motors.

[1:15](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=75s) Um, and about a year ago, uh, my co-founder and I went on stage and we said, uh, at the dawn of the internet era, Bill Gates had written this very famous blog post, and it said, "Content is king."

[1:30](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=90s) Um, and as we are at the dawn of the agentic era, context will be king. Um, since then, it feels like twenty twenty-six is the year of context. Context graphs, anyone?

[1:41](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=101s) Um, uh, you know, every, every two days you see some version of context, uh, popping up. And so what is going on? Um,

[1:51](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=111s) I believe the answer to this kind of is in this reality distortion field that we live in. Uh, I live here in the [REDACTED:location]. Every day or two I have conversations with people which kind of go like, "How far are we from AGI?"

[2:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=125s) And we have a debate, and we're like, "Well, one year, three years," so on. Uh, there is no doubt that the models are getting exponentially smarter by the day.

[2:13](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=133s) Uh, two years ago, they couldn't pass the bar. Today, if they were to take the bar, it was-- they're at the top one percent of test scorers. On the other hand, they're not exponentially more useful by any benchmark.

[2:24](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=144s) Uh, one out of five, you know, AI use cases actually make it to production. Um, you know, fifty-six percent of CEOs say that there's zero financial benefit from AI today.

[2:37](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=157s) So what's going on? I believe hidden in plain sight is actually, um, how performance is measured in the human world. Uh, cognitive intelligence doesn't really determine real-world effectiveness. Uh, in fact, only ten percent of job performance variance is explained by IQ.

[2:54](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=174s) Like, just think about it. Would you say your smartest, um, you know, teammate who scored the highest on the SATs is also your best teammate? Or would you say, "No, it's the person who works the most and takes the most feedback and learns the fastest"?

[3:11](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=191s) In the real world, we care about performance, and performance is outcomes that you deliver in the real world. And performance is a function of two things. It's a function of intelligence, which is cognitive horsepower.

[3:22](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=202s) That's what the model benchmarks measure every day. Uh, but it's also a function of context. This is what they say in the human world as learning on the job, right?

[3:31](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=211s) Knowledge and skills and expertise that you learn over time.

[3:35](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=215s) And in the last decade, uh, we have compounded on one of those parameters. Uh, intelligence has thousand X'd in the last decade. Just in the last six months, we have two X'd on that axis.

[3:48](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=228s) On the other hand, context, the situated knowledge of your business, that's barely moved. We've moved some data to the cloud, uh, but that's about it. It's otherwise locked in dashboards and Slack threads, and, uh, the head of that analyst who might be leaving next week.

[4:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=245s) Um, and so the question ahead of us, and I really believe this is the next frontier, is how do we help AI build context about our business? Um, and every time I'm faced with a question about how do we help AI do this, I always like to go back and understand how did we help humans do this?

[4:25](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=265s) Uh, so I'm gonna take you into the life of, you know, a, a exemplar employee, Maya. Uh, let's say she's a data analyst at McContext Burgers 'cause I thought I was going to be creative, and I'm not very creative.

[4:39](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=279s) Um, and, you know, let's say she's that analyst that everybody, you know, pings in your company, uh, right? She's the person that everybody sends a message to every morning when they're trying to solve a problem.

[4:50](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=290s) So let's say this morning, uh, there's a franchisee owner who sends her a message and says, "Why is my drive-through time up this week? Why is this metric up this week?"

[5:00](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=300s) Sounds like a really simple question, um, but it's actually a really complicated question to answer.

[5:07](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=307s) Just to answer this one very simple question, Maya first needs to know, uh, what is drive-through time, uh, and who's asking. Is it finance or is it, you know, my ops team?

[5:19](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=319s) And it might mean different things. Uh, but not just that. What does this week mean? Is the cutoff period Monday to Sunday? Is it Pacific Time? Is it Eastern Time?

[5:30](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=330s) Uh, that's knowledge. Like, that's facts. That's the map of the business. Um, but not just that. Uh, there's expertise, uh, right? There's, um, you know, a diagnostic playbook. What, what does a great analyst do?

[5:43](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=343s) They know that, you know, quarter three is a season-seasonal quarter because of weather patterns, and they know to go check if the reason there's a spike is because of seasonality.

[5:54](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=354s) They also know that the company launched a product, uh, just that previous quarter, and so they know to check if that's why the root cause analysis failed. Uh, this is expertise and skills that people pick up over time as they learn on the job.

[6:08](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=368s) Uh, and then there's norms, right? Um, there's, you know, persona scoping. Who's asking the question? How do I answer this question? Um-

[6:16](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=376s) And Maya, she's one of those, like, cool people. She nails it. She sends an answer, not just with the answer, but with the why and the root cause, and she finds the reason for it.

[6:27](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=387s) How did Maya learn to do this? Um, she just joined the company a year ago. Um, first Maya, you know, has f- like, she joined, and she got some training, like all of us do, but that's not where any of us learn, right, in our companies.

[6:42](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=402s) How do we learn? We learn because you shadow, like, the best teammate, and then you see why they're doing something, and then you learn from that, and then you make a mistake.

[6:52](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=412s) Who here has learned more from a mistake than anything else? Right? You make a mistake, and then you learn. Uh, you- your manager gives you feedback, and you learn not to do that again.

[7:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=425s) You deal with an edge case, and then you learn from that. That's how all of us humans learn at work.

[7:12](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=432s) And so then the question is: How do you help build the agentic Maya? Uh, and now I wanna walk you through our experiments and learnings as we've built this at Atlan.

[7:23](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=443s) Um, era one, and this was roughly about 18 months ago now, um, we, uh, started on the, the track of bootstrapping agents. Um, and the way we went about it was, and we started this with our customer experience team, uh, and we did this jobs to be done analysis map, right?

[7:42](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=462s) And so we said, "Hey, if you are someone on our customer experience team, what are all the things that you do on a day-to-day basis?" And then we made some hypothesis.

[7:51](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=471s) We, we said, you know, for example, one part of the job is documentation and meeting prep. Uh, we said, "Well, AI could probably do that job pretty well." Uh, and so we built a scaling factor.

[8:02](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=482s) So on the other hand, relationship management is something that our customer experience team does, and we said, "Hmm, that doesn't sound like something AI is going to be able to do anytime soon."

[8:11](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=491s) And so we built a scaling factor. And then we basically started bootstrapping these individual agents that were, like, built for that specific topic. Uh, our team got creative, so we had Hermione, who is our health intelligence lead, and then we had, you know, Moneypenny, who was our financial risk analyst, and we just made that particular agent

[8:35](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=515s) really good at doing that one thing. Um, and that worked for some time. Um,

[8:42](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=522s) but then we realized there were some challenges with this approach. The first, context engineering. Uh, we got to the point by middle of last year where building an agent was really easy.

[8:52](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=532s) It took, like, five minutes. Uh, but giving it the business context that it took to actually get it to be accurate took forever. Uh, quality of the agent often depended, uh, on the quality of context engineering, and that led to a lot of weird lost trust cases with our stakeholders.

[9:11](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=551s) Uh, then as we started taking this into production, we started seeing that these agents basically were kind of like living on their own island. Uh, now imagine, for example, if you're in a human team, and your marketing changes positioning on your, you know, uh...

[9:26](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=566s) And then they come to the town hall, and they tell you that they changed positioning. And so then, you know, the SDR on your team or your sales development rep, they know that they should use that new positioning.

[9:36](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=576s) This is like the infrastructure that we've built for humans inside our organizations. Agents didn't have that. So our marketing team had these agents, and they started making changes to that, and then our SDR agent on our website was still pitching the old version.

[9:51](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=591s) Uh, we had no idea how any of these things were even connected, so we didn't even know how to, like, run this as, as a team of agents. Uh, when an agent gets something wrong, this is hard.

[10:02](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=602s) Uh, it was really hard to, like, trace back what happened. Was it the model? Was it the agent? Was it the context? Like, where, how do we even go back and fix this?

[10:10](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=610s) Um, and over time, we started dealing with, uh, context sprawl. Uh, we had... The, the, the hard part about this was agents all had their own memory systems to a certain extent, so they were learning.

[10:24](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=624s) They were all learning separately, and they were learning differently. Uh, it became very, very difficult very quickly to say, "Okay, what does the single version of truth here look like?"

[10:34](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=634s) Um, and then over time, we actually went through, in the last 12 months, we've gone through cycles of, at the agentic layer, about 12 months ago, we were using one of these no-code type builders, uh, called Relevance.

[10:47](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=647s) We went from there into Google ADK, then we tried Glean. Uh, start of this year we moved to Claude Code. Now we're kinda like 50/50 Claude and Codex. Uh, and every single time as these changes happened, uh, our context got trapped in each of these individual systems.

[11:04](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=664s) Um, so start of this year, as general purpose agents started to become a thing, we said, "What if there was a different approach with general purpose agents?"

[11:15](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=675s) Uh, again, going back to the human world, well, Maya, she's not an individual star. She's part of a team, right? And, you know, you talk about these dream teams, like Maya and someone who runs customer support and someone who launches ads.

[11:28](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=688s) These people work really well together, and often these dream teams are built on shared context, right? Uh, they have a shared language. Uh, they have a shared picture of what's true today.

[11:40](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=700s) They have shared playbooks. Uh, they have shared norms, who's allowed to make what decision. Uh, and then they learn together. I think this is the most important part of it.

[11:50](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=710s) They have compounding learning loops of what good looks like, uh, and they have shared memory that, you know, "Oh, we launched this thing last quarter, and it, like, was terrible, and we're not going to make that mistake again," right?

[12:01](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=721s) And so we said, "Is there a way to bring that into the way we think about AI in our companies?" And so the mental model we started working on was we said, "Okay, we have these teams of humans, and they're across the board, and can these people essentially start building domain skills?"

[12:19](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=739s) So each of them is responsible for a certain set of skills. All of this goes into this common one place, which is this one- Company brain of sorts, right?

[12:30](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=750s) I like to think of this as the context layer. Uh, and then this has a bunch of retrieval mechanisms, which then talks to the general purpose agent across the ecosystem.

[12:41](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=761s) So then we started a experiment. Uh, this is some version of what our marketing team ended up building. So you'll see on the left, those are all the systems that our marketing team uses.

[12:52](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=772s) So data systems, our social and community platforms, our ad platforms, our analytics platforms. Um, and then you'll see this agent block. Uh, we build this very specifically for, um, having openness.

[13:08](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=788s) So we had Claude Code and Cowork. We also had our own Claude that we deployed, which has, you know, essentially talks in our Slack channels. Um, and then we used some external products, like Qualified and Artisan.

[13:21](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=801s) Uh, in the middle is kind of this context layer that our team started building. So think of it as our best SEO person was building their SEO skill. Uh, our best competitive intel person was building the best competitive intel skill, and that kind of became this common repo that we were building into and pulling out from.

[13:44](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=824s) This sort of became our living brain. Over time, we realized there were some things that we needed in this brain, right? Uh, we realized we needed a data graph, like if, for example, our autonomous ads agent, we realized it needs to do analysis on a daily basis, so, like, which table should I go pull from?

[14:00](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=840s) Uh, we needed a library of skills. We also needed some other things, semantics, metrics, what is ARR, how do you measure that? Uh, what is a qualified lead in our company?

[14:11](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=851s) Uh, and org structure, entities, things like that.

[14:16](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=856s) Over the last six months, we ended up creating about three hundred skills and forty agents in this team, uh, which has been incredible. Uh, but then with this approach too, we realized that there were some challenges.

[14:29](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=869s) We realized that context kind of needs to be managed like code. Um, so some challenges, let's pick skills. Uh, dependency management became really complicated. So, for example, we have this competitive intelligence skill, and it learns from the market on what's changing in the market, and it improves.

[14:49](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=889s) Um, it feeds our category positioning skill, which then feeds our sales battle card skill. Uh, now each of these skills is learning and evolving, uh, but every time they learn and evolve, it breaks something downstream.

[15:03](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=903s) Uh, and these skills very quickly start getting outdated and start drifting. Uh, who owns skill quality became another thing, like who eventually owns the quality of this? Security and governance was a nightmare.

[15:16](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=916s) Uh, we had secrets hardcoded in .env files. Uh, it, it was people were downloading these public skill repos. This-- The whole thing was like a nightmare. Um, and then I talked about context portability across all these multi-agent systems.

[15:31](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=931s) I started this talk by saying WTF is a context layer. Uh, these are the problems that a context layer is meant to solve. Um, the question I like to ask is: What does the GitHub for context look like?

[15:44](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=944s) Um, few thoughts. Uh, company context needs lifecycle management, collaboration, and versioning, uh, just like code does. Uh,

[15:56](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=956s) you know, there's questions like what's local context, what's global context, how do I keep this updated, so on. Uh, some thoughts in this, can skills have a PR flow just like code does?

[16:08](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=968s) Uh, can that have a self-learning, learning loop that's baked into it? Uh, what does quality management look like? Can you have security and postures-- posture management associated with that?

[16:20](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=980s) That's really, like, the first step. Uh, I see this as, like, having something that has built-in versioning and quality and dependency management. So you should be able to say, "Hey, this thing impacts all these other things.

[16:31](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=991s) This is the approver. This is the maintainer. These are the contributors." How do you build, like, kind of human plus AI workspaces that these, that these skills, uh, are managed via?

[16:43](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1003s) Second thing, every AI interaction creates more context, and harnessing this, uh, is gold. Uh, uh, there's been, I know, a lot of talks about self-improving loops. Uh, we have found that with traces, deploying a specific harness that actually is specialized in being able to go and reverse construct from that.

[17:02](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1022s) So think of it as AI that's reading through all your traces and almost brings it back to your maintainer loop and says, "Approve, reject, approve, reject. Improve this over time."

[17:13](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1033s) Uh, that's the compounding learning loop. And the third, often a lot of people ask me this question, which is like, "How do I start? Because my business is re-really disparate, and I have all these, like, sixty systems, and how do I even start?"

[17:27](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1047s) One of the biggest learnings we've had is context is hidden in these-- in business systems. Uh, and across this context, quality can really compound. So, for example, if you're able to connect your Salesforce and your HubSpot to your data warehouse, to your application layer, and then you're able to reverse construct how these things are actually connected one

[17:46](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1066s) to another, context today gets lost in every one of those hops. But if you can reverse construct that and then deploy AI on top of it, we've seen incredible accuracy in being able to reverse construct the first version of your company brain.

[18:05](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1085s) So I'll end with this. Uh, the way I think about a context layer is it's a system that turns knowledge and expertise and norms that we talked about, that Maya knows, into a machine-usable context for AI systems.

[18:20](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1100s) Uh, at a very high level, the way I like to think of it is it looks like this. Uh, it continually is mining context from your business systems. It's feeding this into that one company brain.

[18:35](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1115s) It's harnessing this in skills and context development life cycles as your teams go and deploy these agents, and then it has a bunch of ways you can retrieve it.

[18:45](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1125s) So MCP, SQL, Vector retrieval, hybrid assembly, all these different ways that you retrieve it and pull back from traces and build this compounding learning loop. [sniffs]

[18:58](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1138s) Today, we're largely building agents by hard coding context. [laughs]

[19:07](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1147s) The scale of this problem, I truly believe, is underhyped because with scale, this can become really unsustainable, uh, and a little dangerous. Like all of us know this, this old joke, which is if you ask sales and finance the revenue number, you're going to get two different numbers.

[19:22](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1162s) Uh, we're fast approaching a moment of starting to deploy autonomous systems where the same thing is starting to happen.

[19:31](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1171s) So I'll end with one last thing. I started this presentation by saying context is king. Um, I'd like to end it by saying context is also IP. Something I think a lot about is in a world where you and your competitor have access to the same models and the same intelligence, what differentiates a company?

[19:52](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1192s) What differentiates a customer support agent at American Express versus Amazon? Uh, that's how you do business. That's what makes your company special. Uh, context is how we take and encode our culture and our norms into something that we will be proud of as we build autonomous frontier firms, um, and

[20:17](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1217s) that's all I had. Uh, you can find me [REDACTED:username] on Twitter, um, or write to me. We are actively working with folks on the frontier on going and shipping and building company brains, um, so if you'd like to talk to us, feel free to reach out.

[20:33](https://www.youtube.com/watch?v=8G_1-3IO4ZQ&t=1233s) Thank you. [audience applauding] [outro jingle]
