AI Engineer World's Fair 2026

AI-Native Organisations Run on Skills: How to Structure and Scale Them — Imad Touil, QuantumBlack

Read the talk

AI-Native Organisations Run on Skills: How to Structure and Scale Them

Imad Touil explains how reusable agent procedures become organizational infrastructure: specialized skills, shared catalogs, model-aware testing, and human owners who keep the whole system useful as it grows.

From a talk by Imad Touil

At a glance

Ideas worth remembering

  • The coding-agent sequence of specify, plan, task, and implement builds a product increment inside a larger lifecycle of strategy, discovery, data preparation, operations, and improvement.

  • Specialized, discoverable skills let teams reuse procedures and compose them into workflows. Progressive disclosure reduces context cost by loading relevant knowledge when needed.

  • A shared registry needs metadata, dependencies, versions, access control, evaluation, and observability—and named people responsible for maintaining its domains.

  • Skills can contain executable scripts and sensitive business logic. Security checks and retrieval permissions address different risks.

  • Retest skills against their tasks and changing models. Automatic evolution increases the need for governance because it accelerates changes to shared procedures.

Creating a skill is easier than maintaining it together

The hands drop as the questions get harder. Who has created and used skills? Who shares them with a team? Who governs and maintains them across an organization? Imad Touil, a distinguished engineer at QuantumBlack, sees only a few hands left for the last question. That gap sets up the practical problem: a useful procedure on one engineer’s machine still needs a way to become shared, maintained organizational knowledge.

The agentic software stack has two loops in Touil’s model. The inner loop is the coding agent and its harness—the runtime that manages context, tools and MCP connections, memory and state, and skill loading. The outer loop is the workflow that arranges skills, sub-agents, MCP servers, and hooks around the work. These layers serve different purposes: the harness supplies execution machinery, while the workflow shapes how that machinery behaves for a particular job.

Supporting infrastructure makes those loops usable across an organization. A sandbox supplies the execution environment. An MCP gateway manages tool access, and a model gateway manages and optimizes access to local or frontier models. A knowledge graph represents core IT systems, code, and shared registries. The context layer then brings the task’s ingredients together: project instructions, tool schemas that explain what can be called, conversation history, and retrieved files or code. Skills operate inside this larger system; they do not replace its tools, runtime, or context.

0:130:43
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

0:13 · section reference included

Specify, plan, task, implement builds one product increment

The familiar coding-agent sequence is straightforward: specify what to build, plan the design, break the work into tasks, and implement them. At organizational scale, that sequence builds a product increment. It sits inside a much longer journey from deciding what the business needs to operating something customers can use.

Product strategy comes first: define success metrics and a roadmap, using market research, competitive analysis, and customer interviews as inputs. Discovery turns that direction into problem statements, possible solutions, validation, experiments, and user stories. Those steps determine whether the proposed implementation is worth doing before a coding agent starts producing it.

Even then, development may depend on data work. A team might need to clean its data catalog, adjust endpoints or integrations with core systems, build a data pipeline, and validate data quality. Preparing those assets makes the product increment possible. After implementation, platform engineering supplies infrastructure and infrastructure-as-code modules; launch starts another phase of performance optimization and incident resolution, feeding the next cycle.

There is also more than one software development lifecycle within a company. Mobile applications, internal employee platforms, customer-facing products, and different departments can require different workflows. A universal workflow would have to absorb all those differences. Touil instead treats the organization as a collection of lifecycles, each with procedures that can be made explicit and shared where appropriate.

2:453:16
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

2:45 · section reference included

Skills carry the procedure that tools alone do not supply

Within the workflow, Touil assigns four components distinct jobs:

  • Hooks: Trigger actions at events along the workflow.
  • MCP servers: Expose tools, often supplied by products the organization already uses.
  • Sub-agents: Execute delegated tasks in separate contexts, reducing pressure on the main context window.
  • Skills: Carry the organization’s know-how about how a task should be performed.

This division explains why connecting tools and delegating work can still leave a workflow underspecified. The agent needs the procedure as well as the means to act.

Workflows are “harness blueprints”: they shape the coding harness’s behavior at runtime. Skills make those blueprints more explicit by supplying repeatable task instructions and, where needed, scripts. Touil describes the intended result as deterministic. The supported mechanism is narrower: reusable instructions constrain the work, and a saved script can perform a specific operation consistently; neither establishes that every model decision or complete workflow result will be identical.

The adoption story moves from Anthropic’s introduction of skills to an open standard and uptake across agent harnesses. Touil’s snapshot of public repositories and registries illustrates growing creation, while his benchmark discussion reports better software-engineering and cybersecurity task outcomes with skills than without them. The comparison supplies motivation rather than an effect size: no scores or experimental conditions are given here, so it cannot tell a team how much improvement to expect on its own tasks.

5:466:17
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

5:46 · section reference included

Borrow the design discipline of microservices

The engineering problem looks familiar from microservices: many small capabilities need to remain understandable and useful together. A skill should describe one specialized task rather than grow into a monolith of everything an agent might need. Smaller responsibilities make it easier to reuse the procedure and combine it with other procedures without duplicating or conflicting instructions.

Several design properties work together:

  • Reuse and modularity: Package a task so another team can use it without recreating the same procedure.
  • Discovery: Make the skill findable by the team or agent that needs it.
  • Portability: Use the shared standard to carry skills across workflows and supporting harnesses; Touil uses moving from Claude Code to Cursor as his example.
  • Composition: Design responsibilities so multiple skills can run together without overlapping rules that fight each other.
  • Consistency: Preserve a repeatable method for performing the task instead of asking every engineer to steer the agent from scratch.

Cost efficiency comes from progressive disclosure: bring in the relevant skill, in the amount needed, at the point when the task needs it. Loading an organization’s whole procedural library into every context would consume tokens regardless of relevance. Selective loading makes the library available without making every task carry all of it. Touil’s aim is to make organizational know-how executable and portable while reducing the context cost of using it.

A data-retention policy makes this concrete. When an agent manipulates customer data, it needs instructions about the applicable retention rules. That procedure is specialized enough to become one skill. It can then be reused wherever customer data is handled, rather than being rediscovered through a new conversation for each application.

8:248:54
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

8:24 · section reference included

Compose retention rules into a regulatory review

The retention-policy example expands into a regulatory disclosure review. Retention is only one responsibility. The catalog also contains disclosure standards, GDPR rules, and fill-in templates. A workflow pulls these skills at runtime so that a feature handling data can be reviewed against the relevant procedures across web, mobile, and other applications.

The visible change is in what the workflow produces: a storable audit report and identified improvements that can feed back into the codebase. The causal sequence is cataloged rules, runtime composition, review, then actionable output. The report gives the review a persistent result; the improvement findings connect it to development rather than leaving it as a separate compliance exercise.

How do separate policy skills become one review with a feedback path? The diagram shows the shared workflow as the meeting point for four specialized inputs. Its two outputs have different destinations: preserve the report, and use the findings to improve the code. Composition lets the rules stay separately maintained while contributing to the same task.

How it fits togetherFrom policy skills to audit output

Instructions for handling customer data under retention rules.

Four specialized skills feed the regulatory disclosure review; the review produces a report and findings that return to development.

9:5410:25
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

10:25 · section reference included

Ungoverned skills accumulate familiar technical debt

Once teams build skills independently, reuse can fail before anyone notices. Teams on the same stack and infrastructure recreate the same procedures. Meanwhile, a skill that worked for its original task can age as models change. Maintaining quality therefore requires testing against both the intended task and newer models, rather than treating the first successful run as permanent validation.

Discovery and ownership are linked. Touil invokes Backstage and the internal developer portal: a service catalog lets an engineer find who owns a microservice without asking around. A skills catalog needs the same property. Finding the procedure is useful; finding the person responsible for fixing and maintaining it makes reuse sustainable. Composition also needs coordination over task responsibilities and domains, because publishing several skills does not automatically make them work together.

Security has two separate mechanisms to address:

  • Imported instructions and scripts: Public skills can contain prompt injection or insecure scripts. Because a skill can execute a script for a task, bringing one into the environment requires a checking pipeline for executable content as well as instructions.
  • Sensitive business logic: Some internal skills encode knowledge that everyone in the organization should not access. Access control must govern who can retrieve and use those procedures.
11:2111:51
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

11:21 · section reference included

Build the catalog, then assign people to maintain its domains

Adoption starts locally. Individuals need an agreed, structured way to create, test, improve, and use skills. Sharing with the team comes next, allowing people working on the same products and stack to refine the procedure together. The centralized platform then makes that work available beyond the originating team.

The platform needs several capabilities with distinct jobs:

  • Searchable catalog and metadata: Help agents and engineers discover relevant skills. Touil proposes an MCP interface for catalog search and a CLI for pulling skills into an IDE or sandbox.
  • Dependencies: Record relationships between skills so the platform can account for what a procedure needs.
  • Versioning and lifecycle: Identify current releases and let an agent detect and retrieve an updated skill while building a feature.
  • Access control: Make it possible to understand and govern who accesses which skills.
  • Evaluation and observability: Keep quality assessment and operational visibility alongside distribution.

The hard question arrives after those capabilities exist: who governs them? Architecture leads, infrastructure leads, and cybersecurity leads need to own appropriate domains and ensure updates follow organizational policies. The precise assignment depends on the company’s structure. A platform can distribute a release; people must decide what belongs in it and what changes are acceptable. This is where, in Touil’s phrasing, technology stops solving the problem.

What makes reuse become shared improvement rather than one-way downloading? The circulation below connects a maintained catalog to local execution and contributions back. Teams pull shared skills, use them, and return improvements. Named domain owners provide the policy responsibility around that circulation.

How it fits togetherShared skills circulate through teams

Searchable metadata, dependencies, versions, access control, evaluation, and observability.

The catalog distributes procedures to local environments, while improvements return for shared use under human domain ownership.

13:1313:44
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

13:13 · section reference included

Make existing work visible before building it again

Touil illustrates the organizational effect with a simulation of fifteen teams over six months. Its inputs include engineers’ skill contributions, average daily utilization, duplication between teams, and quality and security ratios. It illustrates uneven productivity, cost, and implementation quality under different maintenance practices; it is a constructed scenario rather than measured evidence of a six-month organizational deployment.

The regulatory example explains the cost mechanism behind the scenario. Without a maintained procedure, a human repeatedly steers the agent toward the required rules—“vibe coding back and forth.” Each correction consumes time and tokens, and the outcome depends on that person’s knowledge. A reusable regulation skill supplies the procedure upfront, reducing the need to reconstruct it during every implementation.

The governed scenario changes what happens when the next engineer tries to create a skill. Once a shared skill is published, the coding harness can discover that it already exists and pull it. Discovery interrupts duplication at the moment it would otherwise begin. Teams gain common procedures even while their work remains different.

The same approach must extend to complete workflows. Skills are only one component. An engineer who needs to provision infrastructure should be able to retrieve a workflow with its required skills, run and test it, and return improvements to the centralized platform. That shares the arrangement of the work as well as the individual procedures it uses.

15:1415:44
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

15:44 · section reference included

Evaluate skills before accelerating their evolution

The closing recommendations begin with a skills registry. Touil points to internal developer portals as a familiar place for centralizing this capability, alongside tools dedicated to skills. The practical requirement is a shared place to discover and maintain procedures; the product choice can follow the organization’s existing platform.

Evaluation remains an open design question. A useful starting point is linting or static checks against Anthropic’s skill best practices, looking for structural and invocation problems. A poorly structured skill may be hard for an agent to use; a skill that is not invoked appropriately cannot improve the intended task. These checks complement the earlier requirement to validate task performance and retest as models change.

Auto-evolving skills raise the stakes. A closed loop can keep changing procedures automatically, but Touil’s response to the promise is blunt: “So what?” The earlier maintenance problems still apply, now with a machine accelerating the changes. Ownership, evaluation, security checks, and organizational policies need to govern that loop. Otherwise, the organization is simply maintaining automatically changing skills without a way to keep their evolution aligned with its needs.

18:4119:11
Suggest correction

This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.

18:41 · section reference included

Resources

Read the complete timestamped transcript
  1. 0:01

    [music]

  2. 0:13

    Thank you for joining me today. My name

  3. 0:16

    is Imatel. I'm a distinguished engineer

  4. 0:18

    at Quantum Black. And today talk I want

  5. 0:21

    to really cover the uh how AI native

  6. 0:24

    organizations runs on skills. But before

  7. 0:26

    I really get started, um I just want to

  8. 0:28

    do like a quick exercise to show of

  9. 0:30

    hands. Can you raise your hand if you

  10. 0:32

    have already created and using skills?

  11. 0:36

    Amazing. Now can you keep your hand up

  12. 0:38

    if you are using it and sharing it

  13. 0:40

    within your teams?

  14. 0:42

    Great. Now keep your hands up if you

  15. 0:45

    have governed maintained skills across

  16. 0:48

    your organization.

  17. 0:50

    Amazing. I see a few hands. But that is

  18. 0:52

    what this talk is about. Today I'm

  19. 0:54

    really looking to break down why this is

  20. 0:56

    really critical, how it's important and

  21. 0:58

    how actually you can adopt it across

  22. 0:59

    your organization. But before I get

  23. 1:01

    started, what I really want to cover is

  24. 1:04

    actually the agentic software stack. So

  25. 1:07

    the agentic software stack has two

  26. 1:08

    loops. So the first loop is all you

  27. 1:10

    know, right? It's the code and agents or

  28. 1:11

    the code agents harness right

  29. 1:14

    at this you have some core components.

  30. 1:16

    You will have your context manager, the

  31. 1:17

    tool and MCPS, memories and states and

  32. 1:20

    skills loader, right? But then there's

  33. 1:22

    an outer loop which is your workflows,

  34. 1:25

    right? Those has skills, sub aents, MCP

  35. 1:29

    servers that you use and some hooks.

  36. 1:31

    Sometimes you need them to have this

  37. 1:33

    running properly. You will need some

  38. 1:35

    enablement components at the bottom. So

  39. 1:37

    what you would have is an environment

  40. 1:39

    sandbox. You would have your MCP gateway

  41. 1:41

    to manage and simplify all of the MCP

  42. 1:43

    tools across your organization. a model

  43. 1:45

    gateway to again manage and optimize for

  44. 1:48

    all of your LMS either like you know

  45. 1:51

    open source like running locally or or

  46. 1:54

    or frontier models and also a graph

  47. 1:57

    knowledge graph that abstracts your IT

  48. 2:00

    core systems your codebase your skills

  49. 2:02

    registry and at the end workflow

  50. 2:04

    marketplace and then you have your

  51. 2:06

    context layer right the context layer

  52. 2:08

    will bring all of what is needed to get

  53. 2:10

    the task done so this is the project

  54. 2:12

    instruction think about it as the cloud

  55. 2:14

    code MD file the agency file your tools

  56. 2:16

    and MCP schema is actually to understand

  57. 2:18

    which tool to use and when your memory

  58. 2:22

    right conversation history with the end

  59. 2:24

    user the human is in the loop and

  60. 2:26

    finally the retrieved contents that's

  61. 2:28

    you can pull from either like files your

  62. 2:30

    codebase etc now what I want to really

  63. 2:33

    focus on today is the workflow right and

  64. 2:36

    I think this is where we kind of like

  65. 2:38

    think um it's kind of a pretty simple um

  66. 2:41

    uh workflow in day-to-day when you're

  67. 2:43

    trying to actually create an end to-end

  68. 2:45

    product software delivery life cycle in

  69. 2:47

    the organization. In reality, we all

  70. 2:49

    have seen the four steps. specify to

  71. 2:52

    define what you want to build the design

  72. 2:53

    or plan to plan you know what what are

  73. 2:56

    you looking to build then you go to the

  74. 2:59

    tasks you break it down into tasks and

  75. 3:01

    then finally you start implementing

  76. 3:03

    right I think this looks familiar I

  77. 3:05

    think this is how most of our coding

  78. 3:07

    agents actually are kind of shaped today

  79. 3:09

    in reality that is not how it is

  80. 3:11

    composed and at scale when you look at

  81. 3:13

    the organization complexity

  82. 3:16

    this is just one step in the journey

  83. 3:18

    this is like building a product

  84. 3:19

    increment

  85. 3:20

    When you really look at the overall end

  86. 3:23

    to end life cycle of something that a

  87. 3:25

    business want to build to capture the

  88. 3:27

    value out of it all the way to ship it

  89. 3:28

    to the client, you will start first of

  90. 3:31

    all by defining your product strategy.

  91. 3:32

    What to build and how to build it,

  92. 3:34

    right? You set the define the success

  93. 3:37

    metrics. You identify and you break down

  94. 3:39

    your plan road map for your products.

  95. 3:42

    And to do this, you may need a lot of

  96. 3:43

    insights, right? So then you do like

  97. 3:44

    your market research. You do a

  98. 3:46

    competitive analysis. You bring this as

  99. 3:48

    an input with some customer interviews.

  100. 3:51

    Then you go to the discovery side,

  101. 3:52

    right? So then you start discovering.

  102. 3:54

    Okay, now I understand what to build.

  103. 3:56

    I'm going to break this down into like

  104. 3:57

    some problem statements, find the

  105. 3:59

    solution, validate the solution, and

  106. 4:01

    then probably like experiment and then

  107. 4:03

    create user stories. And before we start

  108. 4:06

    building in reality actually need to

  109. 4:08

    prepare your data, right? In some cases,

  110. 4:11

    you will need actually to clean up your

  111. 4:13

    data catalog that will support the build

  112. 4:15

    of your products or maybe adjust some of

  113. 4:17

    the endpoint connection and integration

  114. 4:19

    to your core systems that will help you

  115. 4:21

    actually build your products.

  116. 4:24

    And this is where the data product

  117. 4:25

    delivery comes. So you build your data

  118. 4:28

    pipeline, you validate your data quality

  119. 4:30

    and you put your catalog your

  120. 4:33

    [clears throat] catalog data assets

  121. 4:34

    ready for development. Then we kind of

  122. 4:37

    like go back to the product increment.

  123. 4:39

    That is where we will start. But then

  124. 4:41

    building a product in every organization

  125. 4:43

    that I have been uh serving for the past

  126. 4:45

    18 years in my career. I can see that in

  127. 4:48

    one organization you will find like

  128. 4:49

    different DLC iss kind of like uh you

  129. 4:52

    know scatterers across organization.

  130. 4:54

    Some of it is actually for a mobile

  131. 4:55

    application other is a different

  132. 4:57

    department or different platform. Some

  133. 4:59

    of it is internal platform that is for

  134. 5:01

    your um employees. Other is actually

  135. 5:03

    customerf facing. So it is not like a

  136. 5:05

    one workflow that can actually build

  137. 5:06

    anything you want for your organization.

  138. 5:09

    When you figure out like what to build

  139. 5:10

    and how to build it, you need to run it.

  140. 5:12

    Then we come to the platform engineering

  141. 5:13

    ops, right? That is again when you have

  142. 5:15

    your provisioned infrastructure,

  143. 5:17

    thinking about how you you build your

  144. 5:19

    your infrastructure as code modules etc.

  145. 5:22

    And then you launch your products and

  146. 5:24

    the moment you launch it then you start

  147. 5:26

    kind of like the journey of optimizing

  148. 5:28

    the performance of your products trying

  149. 5:30

    to look forward like any incidents to

  150. 5:32

    resolve and then you start the loop

  151. 5:34

    again. Right? So at scale when you look

  152. 5:37

    at really building a digital platform

  153. 5:39

    not like a very simple products that you

  154. 5:41

    can you know solo build and deploy the

  155. 5:44

    landscape is way complex that is

  156. 5:46

    expected and what you're looking at here

  157. 5:47

    is literally like probably 10 20% of

  158. 5:49

    what is it and is really different from

  159. 5:51

    organization to organization now going

  160. 5:53

    back to the uh the stack right when you

  161. 5:57

    look at the workflows there's like four

  162. 5:58

    core components the first one is hooks

  163. 6:02

    MCP servers and sub aents those are kind

  164. 6:04

    of

  165. 6:05

    given but it doesn't really bring the

  166. 6:07

    right kind of like structured value to

  167. 6:09

    your workflows right that's why scales

  168. 6:11

    is one of the critical components in

  169. 6:13

    your workflows hooks basically what it

  170. 6:15

    does it just kind of like pre kind of

  171. 6:17

    like trigon events to to do something

  172. 6:19

    along your workflow the MCP service we

  173. 6:22

    all know that you know you may need an

  174. 6:23

    MCP tool but tell me like who actually

  175. 6:26

    build a lot of MCPS we just use MCPS

  176. 6:28

    tools that is actually provided by the

  177. 6:31

    tool that we used to use before right so

  178. 6:33

    we don't really own

  179. 6:34

    The sub agents is just to minimize the

  180. 6:36

    context window. We just delegate to sub

  181. 6:38

    agent to execute a specific task what is

  182. 6:40

    needed. So at the end of the day you

  183. 6:42

    will find all of your knowhow is

  184. 6:44

    actually at the skills level. And if you

  185. 6:46

    don't have the right structure of your

  186. 6:47

    skills then you're not really having a

  187. 6:49

    deterministic workflow. And one thing to

  188. 6:52

    mention is workflows think about them as

  189. 6:55

    harness blueprints that actually shape

  190. 6:56

    the behavior of your coding harness for

  191. 6:59

    example in the runtime.

  192. 7:03

    Now looking at the rise of skills

  193. 7:05

    adoption right so just eight months ago

  194. 7:08

    entropic published the first article

  195. 7:10

    about skills right two month later we

  196. 7:13

    had this standard an open standard that

  197. 7:15

    is adopted and starts a lot of agent

  198. 7:18

    harnesses starts adopting this new

  199. 7:20

    standard [snorts] around February um

  200. 7:22

    this year we have seen most of the

  201. 7:25

    actually agents adopted this even though

  202. 7:27

    you don't see them right if you pay

  203. 7:29

    attention when the the uh the agent is

  204. 7:31

    thinking And you can see that he's

  205. 7:32

    pulling skills um as he's going and

  206. 7:35

    doing the task. The other implication

  207. 7:37

    here that you need to pay attention to

  208. 7:39

    is actually the number of skills

  209. 7:40

    created. Right? This is just like

  210. 7:41

    literally a snapshot that I did across

  211. 7:43

    some public GitHub repost, some public

  212. 7:45

    skills registries, right? There's way

  213. 7:47

    more than this publicly and within your

  214. 7:50

    organizations, right? So the de the

  215. 7:52

    creation of skills and the demand is

  216. 7:54

    rising. But we need to understand why

  217. 7:58

    in the latest skills bench comparing the

  218. 8:02

    latest models against like you know

  219. 8:04

    running the same task against auto

  220. 8:06

    engineering and cyber security without

  221. 8:08

    skills it did well right because that's

  222. 8:10

    what expected and it's going to continue

  223. 8:12

    to be improving day after day right but

  224. 8:15

    then when we applied skills that's our

  225. 8:17

    more deterministics the outcome was

  226. 8:19

    clearly higher than than That's it.

  227. 8:24

    Now if you think about now like how like

  228. 8:27

    the anatomy of skills and how you

  229. 8:28

    actually design and implement skills in

  230. 8:29

    your organization. This is not like a

  231. 8:31

    new problem that we are trying to solve

  232. 8:33

    here, right? We have solved this with

  233. 8:34

    the microser kind of movements, right?

  234. 8:36

    So your microservices need to have all

  235. 8:38

    of these design principles, right? This

  236. 8:40

    is like a software kind of like you know

  237. 8:41

    problem that we use to solve is is

  238. 8:43

    similar. So skills need to be reusable,

  239. 8:45

    right? Need to be modular, need to be

  240. 8:48

    disco like you can discover your skills.

  241. 8:50

    So if you are sitting in one team and

  242. 8:52

    you need the skills you actually can

  243. 8:53

    automatically discover and capture the

  244. 8:54

    skills it's portable that you can

  245. 8:57

    actually use skills across workflows but

  246. 8:59

    also you can use skills across harnesses

  247. 9:00

    again everyone adopted the same

  248. 9:02

    standard. So if I'm having a skill on

  249. 9:04

    cloud code and I want to move it to

  250. 9:05

    cursor, it's going to just work.

  251. 9:08

    Specialized skills that is where the

  252. 9:09

    value you should not build like a one

  253. 9:11

    skill like a monom or again it should be

  254. 9:13

    specialized to define one tasks

  255. 9:15

    specifically. Composible uh skills need

  256. 9:18

    to be designed in a way that's actually

  257. 9:19

    can compose. So you don't have

  258. 9:20

    duplication across your skills when

  259. 9:22

    you're trying to run them conflicts each

  260. 9:24

    other. Consistence that is actually what

  261. 9:27

    one of the key items of skills is

  262. 9:28

    actually consistency and deterministic.

  263. 9:30

    And finally cost efficience and for cost

  264. 9:32

    I can go for another hour talk but it's

  265. 9:35

    basically the skills comes to solve a

  266. 9:37

    key problem around the context window

  267. 9:39

    right is actually putting with the

  268. 9:41

    disclosure pro progressive disclosure

  269. 9:43

    pattern the right skills the right

  270. 9:45

    amount of skills in the right time to

  271. 9:46

    solve the right problem and that's

  272. 9:48

    reduce the token usage. This define a

  273. 9:51

    new unit right that makes your knowhow

  274. 9:53

    in your organization executable portable

  275. 9:56

    and cheap.

  276. 9:58

    On the right hand side, you just a very

  277. 9:59

    simple example of data retention policy.

  278. 10:01

    Right? When it came to regulation, you

  279. 10:04

    need to understand and make sure to

  280. 10:05

    instruct your agents while you

  281. 10:07

    manipulating. For example, um your

  282. 10:09

    customer data, you should make sure that

  283. 10:12

    this data manipulates according to the

  284. 10:14

    regulation, right? And that bring me to

  285. 10:17

    the next example.

  286. 10:20

    On the left hand side, you can see that

  287. 10:21

    there's like a think about like a a

  288. 10:24

    catalog of skills. And in the right hand

  289. 10:26

    side is your harness and the output is

  290. 10:28

    on the right hand side. Right? So the

  291. 10:30

    composible skills at the regulation

  292. 10:32

    level. You have the skill that I just

  293. 10:34

    showed earlier which is the retention

  294. 10:36

    policy. But you will need disclosure

  295. 10:38

    standards. You will need the GDPR rules

  296. 10:40

    that need to be respected. You will need

  297. 10:42

    the fill in templates. Right? So all of

  298. 10:44

    this kind of think about it is like

  299. 10:45

    defining how any data any feature that

  300. 10:48

    is built across your web, mobile um like

  301. 10:51

    you know different applications across

  302. 10:53

    your organizations is really respecting

  303. 10:55

    these rules and this gets pulled

  304. 10:57

    automatically by on the runtime by the

  305. 10:59

    regulatory disclosure review workflow

  306. 11:02

    and the outcome is expected is

  307. 11:04

    deterministic. You have an audit uh

  308. 11:07

    audit reports that you can actually

  309. 11:09

    store. you have um specific uh

  310. 11:12

    identification of if there's anything to

  311. 11:14

    improve and that this kind of like loop

  312. 11:15

    back to improve your your codebase.

  313. 11:21

    However, if we don't govern skills, we

  314. 11:23

    will start creating a new class of

  315. 11:25

    technical depths, right? First of all,

  316. 11:28

    you will find out that you are having a

  317. 11:29

    lot of duplication in your organization.

  318. 11:31

    So if teams are not collaborating and

  319. 11:33

    everyone is think about it using the

  320. 11:35

    same technology stack the same

  321. 11:37

    infrastructure you're for sure building

  322. 11:39

    the same skills over and over again

  323. 11:41

    without sharing them quality. If you

  324. 11:44

    don't test and make sure that you're

  325. 11:45

    maintaining and you're validating your

  326. 11:46

    skills not against your task but also

  327. 11:49

    against the latest models that comes

  328. 11:50

    right then the quality starts

  329. 11:52

    degragating over time.

  330. 11:56

    You should be able to discover your

  331. 11:57

    skills in in reality without a

  332. 11:59

    governance you cannot really discover

  333. 12:01

    it. Think about it as the the backstage

  334. 12:03

    the IDP right it's come to solve a

  335. 12:05

    problem where okay I need to understand

  336. 12:07

    who owns this microser back in the days

  337. 12:09

    right I don't need to talk to anyone I

  338. 12:10

    just need to tap into the service

  339. 12:12

    catalog and immediately find who

  340. 12:13

    actually own it that is brings ownership

  341. 12:16

    part if you don't have an owner then no

  342. 12:18

    one will be able to maintain scale those

  343. 12:21

    skills

  344. 12:22

    composibility is not something that

  345. 12:24

    comes by default you need to have a

  346. 12:26

    governance way to to align what to build

  347. 12:28

    and how to design it is think about the

  348. 12:30

    domain driven approach that we have been

  349. 12:31

    taken also for for many years, right? Is

  350. 12:33

    similar to how you shape your skills

  351. 12:35

    catalog.

  352. 12:37

    Security again some of uh all of us like

  353. 12:40

    we experiment with the public skills,

  354. 12:43

    right? But when you think about it, some

  355. 12:45

    skills may have some prompts injection

  356. 12:46

    and skills actually does have scripts

  357. 12:48

    because that is the deterministic part

  358. 12:50

    of it because it can run a specific

  359. 12:51

    script for a specific task. So if you

  360. 12:54

    don't have again a pipeline that check

  361. 12:55

    your security, you may be pulling

  362. 12:57

    something that is insecure

  363. 12:59

    and permissions. Not every skills is

  364. 13:01

    actually something that anyone in the

  365. 13:03

    organization should access. Some skills

  366. 13:04

    may have some business logic that is

  367. 13:07

    very sensitive, right? So the access

  368. 13:09

    control is is also crucial at this

  369. 13:11

    stage.

  370. 13:13

    Now how to bring this to your

  371. 13:14

    organization? First of all, you need to

  372. 13:16

    allow at the individual level to create,

  373. 13:19

    test, improve and use those skills.

  374. 13:22

    Again, it shouldn't be random. It should

  375. 13:24

    be structured way. There's a different

  376. 13:25

    tools out there. You just need to decide

  377. 13:27

    which tool actually you want to agree on

  378. 13:28

    and you use that mechanism at the

  379. 13:30

    individual level. The moment you create

  380. 13:32

    a skills, you need to be sharing it with

  381. 13:33

    your team, right? Your team starts

  382. 13:36

    collab collaborating to improve the

  383. 13:38

    skills and think about it. You build the

  384. 13:39

    same technology stack, building the same

  385. 13:41

    products. So, it's going to evolve

  386. 13:42

    really quickly. But then you move on to

  387. 13:44

    a very critical point which is the

  388. 13:47

    centralized platform. That is where all

  389. 13:50

    of what I've been covering so far come

  390. 13:52

    to play. You need a centralized platform

  391. 13:54

    that have a catalog with metadata in it

  392. 13:57

    that's actually can discover skills and

  393. 13:59

    could be searchable. You can have an MCP

  394. 14:00

    that's actually plugged to this cat

  395. 14:02

    catalog, search for the skill and a CLI

  396. 14:04

    to pull the skills back to your either

  397. 14:06

    your IDE if you're locally or to your um

  398. 14:09

    sandbox in your factory. Then you have

  399. 14:11

    the dependencies. So you need to

  400. 14:12

    understand the dependencies between the

  401. 14:14

    skills as well. You have the versioning

  402. 14:16

    uh and and life cycle. So you understand

  403. 14:19

    which version of the skills is actually

  404. 14:20

    the latest. Um and a very good example

  405. 14:23

    when I'm using for example building a

  406. 14:25

    functionality I can the agents

  407. 14:27

    automatically up capture that there is a

  408. 14:29

    latest version of the skill and pull it

  409. 14:31

    right. So this versioning help also to

  410. 14:33

    pull the the the right latest changes

  411. 14:35

    from the skills registry access control

  412. 14:38

    again as I said if you don't know who is

  413. 14:39

    accessing what that is a huge uh uh gap

  414. 14:43

    and finally evaluation observ

  415. 14:44

    observability and then all of this is

  416. 14:47

    actually play around a governance and

  417. 14:49

    this is where technology stop solving

  418. 14:51

    the problem right so you figure out all

  419. 14:52

    of this all good now who's going to

  420. 14:54

    govern this and that is where dep you

  421. 14:58

    know it really depend how your

  422. 14:59

    organization is structured today that is

  423. 15:01

    where you should have your architects,

  424. 15:02

    your engineer leads, infra leads etc and

  425. 15:05

    cyber leads actually sitting down owning

  426. 15:08

    part of those domains and making sure

  427. 15:09

    that the skills we need to get updated

  428. 15:11

    is actually according to the policies

  429. 15:12

    you want to adhere within your

  430. 15:14

    organization and drive this change.

  431. 15:17

    And finally when you get this right what

  432. 15:19

    you will have you would have at the

  433. 15:20

    organization level all of your teams

  434. 15:23

    pulling from one centralized place high

  435. 15:25

    quality skills executing them and

  436. 15:27

    pulling them back um to to the

  437. 15:29

    centralized platform if it is improved.

  438. 15:32

    Now what I want to bring this because

  439. 15:33

    it's a little bit of a inclear view. So

  440. 15:35

    what I created I created a simulation

  441. 15:37

    right. So think about this. This is your

  442. 15:40

    organization today, right? And what I

  443. 15:42

    have here, I have um just a random

  444. 15:45

    teams. I have 15 teams created um five

  445. 15:48

    to 12 like per team. Uh you have skills

  446. 15:52

    per engineers contribution. You have the

  447. 15:54

    average skills utilization kind of like

  448. 15:56

    on average like how much time uh skills

  449. 15:58

    are being pulled per day, the

  450. 16:00

    duplication across the team kind of as a

  451. 16:02

    ratio and the skills quality and

  452. 16:04

    security ratio. Now if I run this across

  453. 16:07

    six months what's really happening and

  454. 16:09

    think about it this is already happening

  455. 16:11

    within your organization is teams are

  456. 16:13

    creating and using the skills right but

  457. 16:16

    we don't have visibility and skills

  458. 16:18

    again they're tightly coupled to your

  459. 16:21

    productivity uplift if for example the

  460. 16:23

    example that I shared earlier on the

  461. 16:25

    regulation if we don't have a skill

  462. 16:27

    about the regulation that is someone is

  463. 16:29

    vip coding back and forth and trying to

  464. 16:30

    figure out exactly how to steer the

  465. 16:32

    agent to implement it properly right

  466. 16:34

    that is burning more tokens from one

  467. 16:36

    side costwise but also the productivity

  468. 16:38

    is spending more time rather than giving

  469. 16:40

    in one shot the right answer and the

  470. 16:42

    quality and security is similar if you

  471. 16:44

    don't have clear you know skills defined

  472. 16:46

    and maintained you will have a low

  473. 16:49

    quality in your implementation because

  474. 16:51

    then it's up to the human to decide this

  475. 16:53

    and different the m in the maturity from

  476. 16:55

    team to team you can see the difference

  477. 16:57

    and that is why for example if I look uh

  478. 16:59

    just randomly at this this is like you

  479. 17:01

    can see the productivity of this team is

  480. 17:03

    a kind of a medium right If I look at

  481. 17:05

    this one is a is a bit of like I don't

  482. 17:07

    know it's low medium productivity

  483. 17:10

    quality and security also medium but

  484. 17:12

    when it came to the cost is really high.

  485. 17:15

    Okay, now let's actually say okay how

  486. 17:18

    this looks like if I governed all of the

  487. 17:21

    my skills in my in my organizations.

  488. 17:24

    What's going to happen is of course some

  489. 17:26

    of them will split right and this is

  490. 17:27

    reality isn't going to be perfect um as

  491. 17:30

    we expect but at least what you will see

  492. 17:32

    you will see actually some common ground

  493. 17:35

    across all of your teams. The moment you

  494. 17:36

    govern, you publish one skill, the next

  495. 17:39

    engineer trying to build a new skill,

  496. 17:41

    the coding agent harness will identify

  497. 17:43

    this skill that is already available and

  498. 17:45

    pull it. Right? So you you almost solve

  499. 17:47

    all of the issues that that I covered

  500. 17:49

    about the governance.

  501. 17:53

    And one last point is when it came to

  502. 17:55

    skills, um it's just one component of

  503. 17:58

    your workflows as I said, right? So that

  504. 18:00

    doesn't mean uh if you figure out skills

  505. 18:02

    that says you're good. No, you need to

  506. 18:04

    apply the same kind of like approach and

  507. 18:07

    solution for your whole workflows. And

  508. 18:09

    you may think to apply this again if you

  509. 18:11

    think about it like if you have a a

  510. 18:13

    centralized platform that have all of

  511. 18:15

    your workflows, right? From one side

  512. 18:16

    you're centralizing the workflows which

  513. 18:18

    is also having the skills but also if

  514. 18:20

    the next engineer came and want to I

  515. 18:22

    don't know like provision infrastructure

  516. 18:24

    they can tap into a workflow and build

  517. 18:26

    that workflow with the required skills

  518. 18:28

    and run it and test it again. And if

  519. 18:30

    it's something you need to be improved

  520. 18:31

    in the workflow, you can easily push it

  521. 18:33

    back um to the centralized platform for

  522. 18:36

    your organization to use.

  523. 18:40

    [snorts] Now before I wrap up, um what I

  524. 18:44

    want to leave you with is this is just

  525. 18:45

    the start the beginning. You see like

  526. 18:47

    this just we're talking about six to

  527. 18:48

    eight months. What's coming next and I

  528. 18:51

    would invite you to already explore is

  529. 18:53

    skills registry, right? You should have

  530. 18:55

    one if not already. And the good news is

  531. 18:57

    all of the players that's been solving

  532. 18:59

    the IDP problem like um internal

  533. 19:01

    developer portal they already start

  534. 19:03

    centralizing this capability right so if

  535. 19:06

    you don't have it today maybe in a

  536. 19:07

    couple of months you will see it coming

  537. 19:08

    but also there's a lot of tools that's

  538. 19:10

    actually solving this specific problem

  539. 19:13

    second is skills eval evaluation there's

  540. 19:15

    still kind of like a discussion on what

  541. 19:17

    is the right approach to to to you know

  542. 19:19

    to to evaluate skills the easy thing

  543. 19:21

    that I found so far very valuable is

  544. 19:23

    actually test like you you l static test

  545. 19:26

    your or evaluate your skills against the

  546. 19:28

    entropic best practices. Right? If the

  547. 19:30

    skill is not invoked properly, if the

  548. 19:32

    skill is not structured properly,

  549. 19:34

    there's a high chance that it's not

  550. 19:35

    going to be high quality. And finally is

  551. 19:37

    auto evolving. And again, this is what

  552. 19:39

    everyone kind of like is is the next

  553. 19:41

    hype right now. Like yeah, I can create

  554. 19:43

    like a an a closed loop that can evolve

  555. 19:45

    automatically my skills. So what right?

  556. 19:48

    If you automatically start this machine,

  557. 19:50

    the impact will be way more than it is

  558. 19:53

    today because what I shared earlier is

  559. 19:55

    going to be just maintaining auto

  560. 19:57

    evolving skills without that governance

  561. 20:00

    in place that actually put the

  562. 20:01

    guardrails for your organization.

  563. 20:03

    And at this point I would leave you

  564. 20:05

    here. Thank you so much for your

  565. 20:07

    listening and looking forward if you

  566. 20:08

    have any question I will be in the

  567. 20:10

    leadership lounge. Feel free to grab me.

  568. 20:12

    Thank you so much. [applause]

  569. 20:28

    >> [music]