AI Engineer World's Fair 2025
The Agent-Native Company
Loading slides…
Read the talk
The Agent-Native Company
When agents do the routine work, employees spend more time directing and reviewing it—and the consequences reach into team structure, hiring, and onboarding.
From a talk by Rick Blalock
Start the morning with your agents
Walk into the office, wave to your coworkers, and sit down. Instead of opening email, check what your agent coworkers completed overnight. Some results are ready to use; others need attention or approval. Your first job is to decide where to intervene and what to send forward. This is Rick Blalock’s opening picture of an agent-native company.
At Agentuity, parts of that picture are already ordinary work. Blalock reports that a request sent to Devin during standup can produce a pull request before the meeting ends. The next human action is review. A discussion about changing a marketing channel can similarly end with someone asking a content marketing agent to make the adjustment.
Blalock also describes seeing an X job posting for a manager of AI workers. Managing agents could itself become a full-time job. But the broader organizational question reaches beyond a new title: what changes when AI becomes central to how work gets done? The progression runs from distinguishing AI assistance from agent-native operations, through team structure and daily work, to the people a company hires.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
What would happen if you removed the agents?
An agent-native company is built around agents that augment human productivity and intelligence. Chat is only one possible interface. The defining scope is broader: AI belongs in the foundation of the product, operations, and culture, where employees routinely depend on it to do their jobs.
A useful diagnostic is the removal test. Take agents out of the workflow and ask which responsibilities return to humans. At Agentuity, code changes in GitHub lead to deployments and releases, after which an agent writes the changelog and documentation. Removing that agent would return an entire recurring task to engineers, not simply remove a convenient writing aid.
Blalock extends the test to costs, throughput, and the product itself. If removing agents makes the company slower and its product less useful, AI has become part of the operating model. His ambitions of making employees 10× or 100× more productive, and helping customers achieve similar gains, are goals rather than measured results.
| Dimension | AI-enhanced | Agent-native |
|---|---|---|
| Role of AI | Occasional assistance | Core operating dependency |
| Without AI | Same model, less efficiency | Routine work and decisions return to humans |
| Driving analogy | Driver assistance | Human-directed autopilot |
| Human focus | Work with extra help | Navigation and higher-level decisions |
The driving comparison makes the division of responsibility explicit: humans still choose the direction, while agents take on routine tasks and smaller decisions. Blalock sees the resulting changes to teams, workflows, and hiring as comparable in scope to earlier industrial transformations, including the automobile age.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Connect departments through agent handoffs
These are early organizational patterns, not a settled template. The first is coverage across departments: product, customer support, and operations each have agents performing routine and important daily work. An isolated AI feature or a single enthusiastic team does not create that arrangement.
The next requirement is interfaces and handoffs between departments. Work has to move across organizational boundaries as well as within them. Turning off those agents would therefore remove both task execution and coordination, leaving people to reconstruct the handoffs manually. For the small Agentuity team Blalock describes, that loss of coordination would be particularly painful.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Human conductors and flatter teams
When people direct agents, their role starts to resemble a conductor’s. That familiar metaphor has a concrete organizational consequence: the hiring profile and reporting structure may need to change. If intelligent systems handle some coordination and execution, fewer layers of middle management may be necessary. A flatter company is a possibility enabled by that redistribution of work, not an automatic consequence of buying agent tools.
Blalock describes a morning product discussion becoming detailed requirements, initial agent implementation, messaging, copy, and documentation by the end of the day. He also reports prototypes taking a few hours, allowing further refinement, learning, and testing. The value of a shorter cycle is that the team can examine more possibilities before committing. As Agentuity grows, he expects its organization to resemble a network of humans and AI more than a conventional pyramid.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Make experimentation compound
Experimentation and iteration are already startup ideals. Agents change how much room there is to practice them: routine work consumes less human attention, and prototype construction becomes another task agents can help with. People can spend more of their time deciding what matters and evaluating what the prototype reveals.
Repeated use can also accumulate useful context. After several months with Cognition’s Devin, Blalock reports that it had documented more of Agentuity’s code and become familiar with how the team handles particular situations. That is an account of growing practical usefulness, not evidence that the underlying model was retrained. Together, agent evolution, human orchestration, and rapid experimentation produce an organization that may not fit an inherited management template—the point of his example of trying to apply a Berkeley MBA framework unchanged.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Morning dispatch, lunchtime review
The daily change is overseeing work that has already begun. Blalock organizes his backlog into theme days, with a mixture of mundane tasks and product ideas that need thought. While walking or driving in the morning, he talks to ChatGPT and uses o3 or deep research to start working through larger questions. He supplies the context: documents, links, prior conversations, and Granola transcripts. By the time he reaches the scheduled task, there is already material to think with.
For bugs and documentation issues reported by users, Linear provides the task context and Devin can begin work that produces pull requests. The historical Linear integration involved ticket scoping and a user launching a first PR pass; the workflow should not be read as every incoming ticket automatically becoming a successful fix. The practical benefit in Blalock’s account is getting work moving before he turns his full attention to it.
The rhythm then becomes:
- Inspect in the morning. Check what agents completed and what still needs attention.
- Dispatch the next work. Give agents the tasks and context they need to proceed.
- Review around lunchtime. Examine the resulting pull requests, collateral, and email-agent outputs.
An employee becomes a lead for the agent counterparts doing parts of that employee’s job. That is management of work and outcomes; it does not necessarily mean becoming a people manager.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Domain expertise with asynchronous execution
Marketing supplies another concrete workflow. Blalock directs multiple content agents to existing copy, asks them to optimize it, and has them determine when social posts should run. The resulting posts are automatically scheduled through Typefully. Human expertise supplies the direction; asynchronous agent work prepares and schedules the material. The account does not specify an approval gate before scheduling.
This combination of domain knowledge and agent direction could reshape job titles as well as daily responsibilities. Blalock suggests roles such as AI customer lead, alongside the already familiar AI engineer. The underlying expectation is that fluency with agents will spread across functions, rather than remain a specialty confined to engineering.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Hire for the ability to guide agents
If employees are expected to guide agents directly, curiosity and adaptability become operating requirements. At Agentuity, Blalock treats them as necessary for using AI in the way the company needs. AI fluency begins to resemble word-processor or keyboard literacy: a basic capacity assumed in office work. His comparison with an earlier era of typewriters and handwriting illustrates how a once-debatable skill can become an ordinary expectation.
Interviews should therefore examine whether a candidate can express their expertise through agents. Can they guide the work, adapt their instructions, and learn the adjustments that make an agent useful? Blalock says unfamiliarity with AI raises warning flags in Agentuity’s interviews, while recognizing that some candidates have had little opportunity to use it. Existing familiarity matters, but so does the ability to learn.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Give new hires an agent setup, not just a job description
The same hiring logic applies to leadership. A VP traditionally brings experience, a network, and the ability to recruit a capable team. In an agent-native company, the evaluation expands: can that leader use agents, and can they recruit other people who can? Once the leader begins building a team, those expectations have to be supported by onboarding.
Blalock proposes two practical changes:
- Engineering support for each team. Attach an engineer who can build the necessary agents and keep them running.
- Agent setup as initial work. Let new hires spend their first few weeks establishing the agents needed for their jobs.
These are proposed onboarding practices. Their logic is that a company cannot expect agent-centered performance while leaving employees without the tools and systems that make it possible.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
A small team’s infrastructure claim
The eventual economic scale remains unknown. Returning to the automobile analogy, Blalock sees agents raising the ceiling on what businesses can do in ways that are difficult to predict. Treating AI as a core business primitive changes roles, skills, culture, and operations together. That creates opportunities, but also friction as existing structures are replaced.
Blalock reports that Agentuity, then about 15 or 16 weeks old with six or seven people, built an entire agentic cloud infrastructure from scratch in a few weeks. He credits internally built agents, Devin, and other tools. This is his account of the team’s delivery, without a supplied architecture breakdown or measured counterfactual for how long the same work would have taken without agents.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Reconsider the experience you bring
Experienced founders face a particular challenge: deciding which lessons still apply. Years of experience can help, but they can also preserve assumptions about how work must be divided, coordinated, and staffed. Blalock urges founders to examine those assumptions critically. He recalls a PwC warning that companies pursuing only small efficiency gains from AI risk falling behind companies changing their operating model more deeply.
The resulting work is organizational: reconsider culture, rewire whole processes, redesign the org chart, redefine roles, and change the skills used to evaluate hires. Blalock’s aspiration is for human–agent teams to scale their impact dramatically, while acknowledging that the transition can be painful. The final question is whether a company is merely using AI—or is ready to be built around it.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
Further reading
Rick Blalock's companion essay on agent-centered organizations, internal automation, and emerging management roles.
Blalock describes turning team conversations into coding tasks and the remaining need to preserve decision context.
- Devin's 2025 release notesDocumentation
Dated records of Devin's Linear integration, codebase intelligence, and human approval controls during 2025.
- Granola 2.0Article
A May 2025 introduction to shared meeting notes and querying team conversations with AI.
Full transcript
Loading transcript…