Pivotal Labs was a preview

The first time I worked at a company that knew what it was doing, my team was embedded at Pivotal Labs in San Francisco. Our employer had hired Pivotal to teach us how to build software. The model was not that Pivotal visited our office. We worked from theirs, paired with their staff every day, and the practices became ours by repetition rather than by training.

I'd come from years of what I now recognize as a cloud of learned helplessness. Confusion about priorities. Frustration with handoffs. The slow drag of teams where nobody quite shared a definition of "done" or "good" or "next." I had assumed this was just what work was. Some places were better, some were worse, but the underlying texture was the same: friction, ambiguity, and small daily failures of coordination that added up to a career of effort that felt heavier than its output.

Pivotal was different. From the first week.

The mechanics were the obvious part. Pairing at every station. Iteration planning on Mondays. Retros on Fridays. The anchor user always in the room. Stories sized by relative effort, not hours. Done meant deployed and validated, not "I think it's working." None of these were exotic individually. What was exotic was that everyone on the team operated under the same paradigm at the same time, without negotiation. The cost of agreement was paid up front, before any of us showed up. We just inherited the operating system.

What that felt like, from the inside, was alignment without effort. Decisions didn't need to be re-litigated. Vocabulary didn't need to be defined. Trade-offs were visible to everyone simultaneously. The team I worked on after I left Pivotal carried that operating system for years, and was the most generative unit I've ever been part of. We shipped at a rate I haven't matched since, with a craft floor higher than anything around us.

Then I spent the next twenty years trying to recreate it. I have not gotten back there.

I've gotten close. I've imported pieces. I've sold the executives. I've trained the teams. I've written the playbooks. Every attempt has hit the same wall: you can't get half a Pivotal. The methodology required a 360-degree wrapper of culture, ritual, environment, and shared training. Any company that already exists has a different wrapper. Trying to graft Pivotal onto an in-flight org is like installing a new operating system while the computer is running. Something always crashed. Usually the ceremonies, because they were the most visible and the easiest to dismiss as overhead.

For a long time I thought the lesson was that the Pivotal magic was unrepeatable. That you had to be embedded, in their building, eating their food, paired at their workstations, taught by their people, to get it. That what I'd lived through was a gift you couldn't carry out the door.

I think now that was the wrong lesson. What I experienced at Pivotal wasn't a gift. It was a preview.

What Pivotal actually proved

Strip the rituals away for a moment. What Pivotal proved was a much simpler claim: small teams of high-judgment people who share a paradigm ship dramatically better than larger teams who don't. The ceremonies were a delivery mechanism for the paradigm. Pairing wasn't valuable because two heads are better than one. Pairing was valuable because it forced two people to externalize their judgment to each other continuously, which compressed the time it took for them to develop a shared model of the problem. The IPM wasn't valuable because Mondays needed structure. The IPM was valuable because it made trade-offs publicly visible before anyone had emotional investment in a particular answer.

Everything Pivotal did was a workaround for the same root problem. It is brutally expensive to get a group of skilled humans to operate under a shared paradigm. Pivotal was willing to pay that cost on behalf of its clients. That was the business. And it worked because they paid it relentlessly, every day, in dozens of small enforcement moves that most companies aren't willing to make.

The cap on the model was the cost itself. Pivotal could not scale the experience past the population of teams willing to be embedded, taught, and rewired. The methodology was sticky once installed, but installation was expensive and slow. Most companies fell off the wagon within a year of disengagement. The IP didn't travel. VMware acquired Pivotal Software for $2.7 billion in 2019 and rebranded the consulting arm as Tanzu Labs. After Broadcom acquired VMware in 2023, Tanzu Labs was shut down in January 2025. Even Pivotal couldn't keep Pivotal alive through the corporate transitions. The model worked. It just didn't compound, and the institution that carried it didn't survive.

The reason I couldn't recreate it for two decades was that I was trying to pay the same cost with weaker tools. Without Pivotal's full enforcement apparatus, every attempt to install the paradigm bled out through the path of least resistance. The pairing got dropped. The anchor user got replaced by a project manager. The shared vocabulary fragmented back into department dialects. The coordination tax was too high for any organization to pay voluntarily once the immune system pushed back.

That tax is the thing that just disappeared.

What changes now

When AI absorbs the coordination layer, the cost of getting a group to operate under a shared paradigm collapses. Not because AI imposes the paradigm. Because AI removes the things that used to fight against it.

Vocabulary fragmentation stops being a problem when the system you're directing forces explicit articulation every time you make a decision. Trade-offs become continuously visible because the agents holding the work surface them as a side effect of doing the work. Handoffs evaporate because there are fewer handoffs. The retro becomes a query against the actual transcript of the week rather than a memory contest between five people who saw five different things.

What used to require Pivotal's whole apparatus to enforce now happens as a consequence of how the work itself is structured.

This is the cap coming off. And it doesn't recreate Pivotal. It does something more interesting. It produces a version of the Pivotal experience that one or two senior practitioners plus an agent system can run, without the enforcement overhead, without the embedded coaching, and without the long install cycle. The team I had after Pivotal, the most generative unit of my career, is now a structure that almost anyone with the judgment for it can stand up. The cost is no longer the gating factor. Judgment is.

That changes who can build a great company, how quickly, and at what size.

Three compressions

To get specific about what changes, three compressions are happening at once.

Case compression. A junior practitioner used to need years of exposure to develop instinct. Patterns came from seeing the same situation play out dozens of times, slowly, across real engagements. Agents collapse that timeline. A junior plus an agent system can run a hundred variants of a real decision in a week, with feedback, and develop calibration that previously took half a career. The bottleneck on judgment was never difficulty. It was bandwidth.

Articulation compression. When a senior practitioner directs an agent, they have to externalize judgment that would otherwise have stayed tacit. The instincts that took years to absorb by osmosis become prompts, overrides, rejected suggestions, written-down reasons. Mentorship used to be capped by the senior's willingness and ability to explain their thinking. Now the explanation is a forced byproduct of getting any work done. Apprentices learn from artifacts that seniors couldn't previously produce.

Stakes compression. Junior work used to be either real and dangerous or fake and worthless. Agents act as a continuous first reviewer, which means juniors can ship real work with a safety layer that didn't exist before. Failure is fast, cheap, and observable. The trust gap that used to take years to close now closes in weeks.

The three compressions together don't eliminate apprenticeship. They redesign it. The new apprenticeship isn't watching a senior over their shoulder for a decade. It's pairing with a senior plus their agent for three or four years, in conditions where every decision is visible, articulated, and reviewable. The transcript is the apprenticeship.

What the new org actually looks like

If you're starting a company today, the structural implication is severe. Don't build the org you would have built in 2018 and bolt AI onto it. That's the path the immune system is pushing you toward, and it produces the companies that look productive and aren't.

Build the org the compressions imply.

That's three or four roles, not three or four hundred. A small senior bench. A paired apprentice layer that learns through structured exposure to senior decision-making. An agent system that handles the execution and surfaces the trade-offs. Almost everyone is essentially fractional. The legal structure is a tax artifact, not an organizing principle.

The career ladder collapses to two categories: building judgment, and exercising judgment. The middle decade of accumulating dues by being reliable in a coordination role goes away. People who would have been remarkable seniors at forty-five are remarkable seniors at thirty-two. People who were going to be coordinators forever lose the role entirely.

What disappears with it: the coordinator middle. The project manager tracking tickets in service of a status meeting. The middle decade of careers built on accumulated context rather than accumulated judgment. The credentials industry that certified coordination capacity will follow. PMP, Scrum Master, Six Sigma, ITIL, and most operations management programs were training systems for a job that's being absorbed. They will linger because credentials always linger. They will not appreciate.

What appreciates: anything that certifies taste, judgment, technical fluency, and the ability to direct intelligent systems. Those credentials largely don't exist in the market yet, which is itself an opportunity, and probably a business.

Companies become small, dense, and high-stakes. Carrying weak judgment becomes very expensive because there are no buffers. Carrying strong judgment becomes very valuable because the throughput per person is high enough that one excellent operator outproduces entire teams from the previous era.

The most generative unit I was ever part of had five people. The version of that unit that's possible now has one or two. That should not feel like a loss. It should feel like a release.

What happens next

The arc plays out in three phases, and the first one is already visible.

Phase one is the AI mandate era. Most companies right now have a mandate: use AI, hit the adoption number, show it in the all-hands. I've written elsewhere about why this fails, but the short version is that the mandate installs AI into an operating model built before AI existed, run by the part of the org most threatened by AI existing. Adoption numbers go up. Output doesn't. We're probably twelve to eighteen months from this breaking publicly at scale, and the post-mandate hangover that follows will be ugly. Procurement budgets will get cut. Champions will get reassigned. A small consensus will form that AI didn't deliver, which will be wrong, but will feel true to anyone whose only exposure to AI was a mandate.

Phase two is the outlier phase. While the mandate companies are tracking adoption metrics, a small number of operations are shipping at output rates that don't make sense by old benchmarks. Small teams, fractional headcounts, revenue per employee numbers that look like typos. These are the companies that rebuilt the operating model rather than bolting AI onto the old one. The first wave is founders, because founders have the freedom to. The second wave is private companies whose boards understand what they're seeing. The third wave is public companies forced to defend their headcount-to-output ratios on earnings calls. By the time the third wave hits, the outliers are no longer outliers. They're the new benchmark.

Phase three is the recognition phase. This is when the question becomes a market. Companies look at what the outliers are doing and ask the only question that's actually load-bearing: what do they have that we don't? It is not the tools. Everyone has the tools. It is not the budget. It is not even the talent in the abstract. The answer, which is uncomfortable for everyone selling something simpler, is that the outliers rebuilt their operating system. The shared paradigm. The judgment layer. The agent coordination spine. The thing Pivotal Labs taught me twenty years ago, now finally portable, now finally affordable, now finally available to be installed without an embedded coaching staff and a four-year culture transplant.

That is the moment Marliin Residency is built for.

Where Marliin lands

I should be direct about my own bias. Marliin Residency is the program I would have wanted access to in every job I had after Pivotal.

We embed senior practitioners with agent teams into companies that want to operate this way. We do the work alongside the client, with the same kind of 360-degree wrapper Pivotal used to provide, except the wrapper is now mostly carried by the agent layer rather than by a coaching staff. The methodology gets encoded into the agents. The agents stay behind when we leave. The next person on the client's team doesn't have to be retrained in the paradigm because the paradigm is running in the system.

This is the part Pivotal couldn't solve, and the reason their model didn't compound through three corporate parents. Four failure modes stacked.

Senior bench bottleneck. Every engagement required a pair of senior humans. Throughput was capped by how many trained Pivots existed and how many places they could be at once.

Methodology fragility. What they installed had no durable residue. It lived in client team memory and rotted on contact with the home org's antibodies. The retro habit got skipped. The pairing got dropped. The shared vocabulary fragmented back into department dialects within months.

Non-compounding IP. Every new client started near zero. The hundredth engagement looked roughly like the first because nothing accumulated outside the people delivering it. There was no asset that got more valuable with each project.

Institutional fragility. Service businesses get stripped when acquirers want margins. Pivotal got stripped twice, the second time terminally.

Marliin's structure addresses each of those directly. Senior bench scales because the agent layer carries throughput, and one principal with a custom agent team produces the output of a small Pivotal team. Methodology has durable residue because it ships as software, and the agents enforce the operating system against the client's antibodies after we leave. IP compounds because every engagement adds to the agent core, and the hundredth client benefits from everything the previous ninety-nine taught us. Institutional risk is lower because the agents stay with the client even if Marliin disappears. Pivotal's clients had nothing when Pivotal went away. Marliin's clients have a working system regardless of what happens to Marliin.

I don't think we're alone in seeing this. I do think we're early. The companies that get this shape right in the next two or three years will outproduce the ones still trying to install AI into a 2018 org chart. If you're founding now, build this way from day one. If you're inheriting an existing org, the work is harder, but the destination is the same.

I have spent twenty years trying to get back to that team I had after Pivotal. For the first time, I can see the road there. It doesn't look like Pivotal. It looks like what Pivotal was a preview of. And it's open to far more people than the gift I once thought I'd received and lost.