Establish the Lab: Hunt Change-Makers and Give Them the Tools Now

You cannot run an FDE team in a company that has nowhere for change-makers to stand. The residency needs a home that is not a CoE, not a vendor war room, and not a Friday hackathon with pizza. It needs a lab: a place where the people who already feel the work can get the tools, the permission, and the harness to ship before the org chart notices.

This is not innovation theatre. Innovation theatre is a glass room, a named programme, and a quarterly showcase that never touches production. A lab, in this regime, is an empowerment system. The speed from idea to execution has never been this short. Talent inside the company has never been this valuable. The companies that win will hunt for the people who can already see the next loop, and give them the stack now, not after a transformation office finishes the operating model.

The gap closed. The org did not.

For most of corporate history the distance between I know what we should do and it is running on Monday was a capital request, a backlog, and a priesthood. That distance is what I wrote about in The Individual Contributor Age and As Fast As You Can Think. Implementation stopped being scarce. Clarity, taste, and domain fidelity did not.

The lab exists because that compression is now structural. An operator with a picture, an agentic IDE, and a safe path to real systems can close a loop in a week that used to be a year's worth of "digital." If you do not give them a place to do it, they will do it anyway, on a personal ChatGPT, a rogue n8n, a contractor's laptop. You will get speed without a harness. That is how you get the next incident, the next shadow process, and the next board slide about "AI governance" that still cannot name a production workflow.

So the lab is not optional culture. It is the only adult answer to a new physics: the cost of not empowering your own people is now higher than the cost of letting them ship.

Your talent has never been this valuable

Models are rented. Domain is not. The person who knows why the exception exists, who the client actually trusts, which field in the CRM is a lie, and what "done" looks like on a Thursday afternoon, that person is the scarce asset. A year ago they were gated by engineering capacity. Today they are a multiplier. Give them a model and a harness and they outrun a programme. Leave them in a queue and you have paid for taste you refuse to use.

This is the part executives still get backwards. They shop for frontier models and named vendors. They under-index the humans already on payroll who have been waiting for permission. The lab's first job is not "explore AI." It is to raise the conversion rate of the talent you already have, operator to shipped loop, idea to artifact, exception to system.

Treat them as interchangeable "users of Copilot" and you will get average output. Hunt the change-makers, the ones who already built the spreadsheet that runs the region, who already automated their own week, who already asked for API access and were told no, and you will get the next operating model for free. They were always there. The tools finally match their speed.

Hunt. Do not wait for volunteers from the town hall.

Change-makers do not raise their hand in the all-hands. They are busy. They are slightly allergic to programmes. They have been burned by innovation labs that wanted a use-case intake form. You hunt them the way you hunt product-market fit: look at who is already pulling.

• Who already ships without a ticket? The brand lead with a private Figma-to-deck loop. The analyst whose Sunday pack is a script. The CS manager who wired Slack to the CRM and never told IT.
• Who is angry in a useful way? Not the cynic. The operator who can name the broken step, the fake field, the approval that lives in email. Anger with a spec is a gift.
• Who do other ICs already ask? Informal technical operators. If three teams route questions through one person, that person is already a lab, just an unofficial one, with no evals and no backup.
• Who stayed close to the work after promotion? Managers who still open the system of record. They can sponsor a residency without becoming a slide owner.

You will not find them with a survey titled "AI champions." You find them by walking the work, by reading the shadow tools, by asking FDE triads who they cannot live without. Then you give those people more surface area, not a committee seat.

What a lab actually is (in this decade)

A lab is a privileged environment for people who will ship into the business. It is closer to a foundry than a research group. It has three jobs, and only three:

• Access, identity, model routes, data products that are allowed, sandboxes that look like production, secrets that are not a Slack paste. If the change-maker still waits six weeks for an account, you do not have a lab. You have a brochure.
• Harness, the same instinct as harness over horse. Interface, memory, permissions, logging, evals, kill switch. The lab does not hand out a chatbot. It hands out a path that can become a running workflow.
• Witness, someone who can say yes on a Tuesday. A lab without a named adult who can unblock data, legal, and security in hours (not quarters) is a waiting room. Empowerment is a latency number.

It is not a place that owns the company's AI narrative. It is not the vendor relationship office. It is not where failed CoE projects go to be rebranded. If the lab's primary artifact is a showcase, shut it down and give the budget to two residencies.

Relation to FDE is explicit. The lab is the center that stays a platform. FDE is the edge that lives in the domain. Change-makers in the business walk into the lab, leave with a golden path, and either ship themselves or pull an FDE triad for the last mile. If the lab tries to do the last mile for twenty business units, it becomes the touring band I already warned you about.

Give them the tools now: not the roadmap to the tools

The tell that an org does not believe the compression: it publishes a 2027 enablement plan while operators are already pasting client data into consumer chat. Speed without a sanctioned stack is not a morality play. It is a forecast. They will use whatever closes the loop.

"Now" means a starter kit you can put in someone's hands this month:

• An agentic IDE on a managed machine, Cursor, Claude Code, or whatever you can actually support, with the company's model route, not a personal API key.
• A skill library they can extend, filesystem-addressable, versioned, boring. Same thesis as the skill library and skills over agent armies. One brand check. One CRM lookup. One eval harness. Do not wait for a platform to be complete.
• A data product that is allowed, a slice of reality with a known classification, not "come back when the lake is governed." Start with the dataset the operator already lives in.
• A path to production that is ugly but real, identity, logs, a human override. Prototype in the lab; graduate into the FDE residency or the operator's own system of record.
• A budget for tokens and time, if they have to steal GPU hours and nights, you have told them this is a hobby. Talent this valuable does not stay in hobbies.

Preferences can wait. Non-negotiables cannot: no secrets in prompts that leave the tenancy, no unlogged production writes, no model that cannot be named in an audit. Put those in the path. Everything else is the change-maker's call. That is the same split as governance the FDE carries rather than waits on.

Empowerment is permission plus a floor, not a speech

People are not empowered by a values poster about experimentation. They are empowered when the expected cost of trying is lower than the cost of asking. The lab sets that floor.

Permission: you may change a named workflow inside this envelope. You may ship a skill. You may fail a loop in public as long as the kill switch works. You will not be punished for a prototype that did not graduate, only for a prototype that pretended it was production.

Floor: you will not be left alone with a model and a prayer. There is a golden path, a human who has done this, evals for the obvious failure, and an FDE you can pull when the last mile is integration. Empowerment without a floor is how juniors get hurt and how legal shuts the lab in month four.

The corporate impotence cycle is leadership consuming summaries of work they no longer touch. The lab is the opposite geometry: executives who want transformation should walk the lab the way they used to walk the floor. If they only see the deck, they will fund the theatre.

Design it like a foundry, not a campus

Physical space is optional. Some of the best labs are a channel, a repo, and a Tuesday desk in the business. What is not optional is density. Change-makers need to see each other's loops. Isolation produces twelve snowflakes. Density produces a library.

Staffing is small on purpose. A workable seed:

• One lab lead measured on loops graduated, not on headcount or vendor coverage. They are a platform person with taste, not a programme manager.
• One harness / platform engineer, identity, routes, evals, logging. The same human (or pod) that makes FDE residencies boring to ship.
• Rotating operators, four to eight change-makers with a time-boxed tour and a named outcome in their real job. They do not "go to the lab." They bring the work in.
• A legal/security partner on a short leash, hours of latency, not a standing committee. If they only attend steering, you have already lost.

Do not staff a research bench unless you are actually doing research. Most enterprises are not. They are trying to install known patterns into stubborn workflows. That is foundry work. Hire accordingly. The micro-team is the unit: small, armed, done when the loop exists.

Career path matters here too. The lab cannot be a parking lot for people the product org did not want. It also cannot be a dead end for the best operators. Sequence: operator-in-residence → technical operator back in the business → FDE or platform. You are growing the IC the org actually needed, same as a well-designed FDE career path.

What graduates. What dies.

If everything lives forever in the lab, the lab is a hospice. Every loop gets one of three labels, in writing, by a date:

• Graduate, running in the operator's workflow, owner named, evals on, FDE or platform holding the harness. This is a win. Celebrate the production path, not the demo day.
• Library, did not become a workflow but left a skill, a connector, an eval, a dataset definition. Reuse is the point. Kill the UI; keep the bone.
• Sunset, wrong problem, wrong owner, or unsafe. Delete it where people can see the deletion. A lab that cannot kill is a CoE.

Time-box like a residency: ninety days is plenty to know. The fractional digital employee test still applies: did you install a capability that keeps working when the original human goes on leave?

How you know the lab is real

Stop counting "experiments." Measure empowerment as an operating system:

• Time from idea to first production-shaped path, not a slide, not a chatbot toy. An operator using it on a real case.
• Change-maker conversion, of the people you hunted, how many shipped a loop that someone else could run? If the lab only produces the lab lead's work, you hired a hero, not a foundry.
• Pull, not push, inbound from the business. Operators asking for a slot. If you are still hunting after six months, either the tools are wrong or permission is fake.
• Reuse, later loops starting from the library. Zero reuse means artisanal theatre.
• Shadow IT down, consumer tools and rogue automations should fall as the sanctioned path gets faster. If shadow rises while the lab is "busy," the lab is slower than the street. That is a product failure.
• FDE leverage, residencies that start from lab skills instead of from zero. The lab and FDE are one system. Score them that way.

Token spend is a cost, not a KPI. Utilization is how you kill a foundry. Hours in the lab that do not graduate are inventory. Inventory should hurt.

The failure modes (you already have these too)

Innovation theatre. Glass walls, a named lab, a waiting list, no path to production. The change-makers visit once, learn the real work still needs a ticket, and never return.

Lab as CoE. The lab owns the narrative, the vendors, and the steering pack. Business units attend. Nothing ships that the business can keep. You reconstituted the monolithic product org and called it curiosity.

Tools as a future tense. Roadmap for access. Meanwhile the talented people are on personal accounts. You taught them the company is the bottleneck.

Open door, no hunt. "Anyone can join." The people who show up are the people with spare time, not the people with the picture. Average inputs, average loops.

Research cosplay. Papers, eval leaderboards, model bake-offs, zero workflow change. Frontier shopping as a substitute for standing next to the work.

No kill switch, then a ban. Someone ships something unsafe. Legal panics. The lab is frozen. The change-makers go back underground. You had a chance to put governance in the path and chose a poster instead.

If you recognize more than one, do not add headcount. Narrow the mission to access, harness, witness. Hunt four operators. Graduate two loops. Then talk about scale.

How to stand it up this quarter

You do not need a two-year target operating model. You need a foundry that exists on a calendar:

Week one. Name the lab lead. Name the security partner with hour-scale latency. Write the non-negotiables on one page. Pick the first data product that is allowed. Stand up the model route.

Week two–three. Hunt. Sit with two business units that already hurt. Identify four change-makers by watching the work, not by emailing a form. Offer them tools and a ninety-day outcome intheir language: the pack, the exception, the campaign lockup, the onboarding path.

Day thirty. First production-shaped path, even if it is ugly. Logging on. Kill switch tested. Operator using it. If you cannot get here, your access story is a lie, fix that before you hire.

Day ninety. Two loops labeled graduate / library / sunset. Skills that a later FDE residency can pick up. A written list of what the platform must make boring next. Only then do you add people.

Vendors will sell you a lab-in-a-box. Sometimes the box is how you get identity and models faster. It is not a substitute for hunting humans. If the only people who can see the work are outside the company, you rented a demo with better lighting.

Lab, FDE, operator: one system

Do not run these as rival programmes. The operator holds the picture. The lab makes it cheap and safe to try. The FDE installs the last mile when trying is not enough. Platform sits behind both. That is the enterprise version of the triad, zoomed out.

Companies that treat the lab as a campus and FDE as a SWAT team will get two theatres. Companies that treat both as how you convert irreplaceable talent into running systems will look, from the outside, like they "adopted AI fast." They did not. They stopped rationing their own people.

The idea-to-execution gap has collapsed. Your change-makers already know what to do with that. Establish the lab so they can do it in the light: tools this month, permission with a floor, a harness that can graduate, a hunt for the people who were already pulling. Talent has never been this valuable. Acting like you have time to wait for the operating model is how you donate them to the street, or to a competitor who handed them Cursor on day one.

Related: Charter the Lab: When to Fund It, When to Kill It, What Legal Signs, The FDE Frontier: Designing Forward Deployed Teams That Compound, and The Lab Next to the Line.