GitLab's Act 2 - The Agentic Restructure Goes Public

Bill Staples published a letter on May 11 that I think will be referenced for years. GitLab, a public company at a $1B+ ARR run rate, has just put its entire operating model into the centrifuge: thirty percent fewer countries, three layers of management removed in some functions, nearly sixty smaller empowered R&D teams replacing the previous structure, internal processes rewired with agents, the long-standing CREDIT values framework retired in favour of three new operating principles, and a generational rebuild of the underlying infrastructure including Git itself. They are not waiting for the agentic era to arrive. They are operationally betting the company on having to be already inside it by June 1.

I have been writing about this transition for the last six months across a series of articles - the empowerment thesis, vibe coding in prod as a management problem, the skills framework, the three-tier AI architecture, the last bottleneck. Staples’s letter is, in a single document, the most public, most concrete version of every argument in that series. Reading it felt like watching a CEO read your draft and turn it into a balance-sheet decision.

This is what the agentic restructure actually looks like when a public company does it openly. It is worth reading the letter cover to cover. This piece is my synthesis: what GitLab did, what it tells us about where the rest of the market is going, and what it validates - and what it risks - about the empowered-team playbook applied at this scale.

What GitLab Actually Announced

The letter is split between an operational reorganisation and a strategic thesis. They are intentionally separated, even though the press will conflate them. Both are worth understanding on their own terms.

The operational moves

1. Country footprint cut by up to 30% where teams are sub-scale. Customers in those markets will be served by partners.
2. Up to three layers of management removed in some functions, because “eight layers is too deep for a company our size and management layers are slowing us down.”
3. R&D rebuilt into roughly 60 smaller, empowered teams with end-to-end ownership. This nearly doubles the number of independent teams.
4. Internal processes rewired with AI agents, automating reviews, approvals, and handoffs.
5. A transparent restructure process with a voluntary separation window through May 18, finalised on or before June 1.

The strategic thesis: ten core beliefs

The world they are building for:

1. Software will be built by machines, directed by people. Agents plan, code, review, deploy, and repair. Humans own architecture, customer problem understanding, and the tradeoffs that require taste.
2. The agentic era multiplies demand for software. Developer platform spend is moving from tens of dollars per user per month to hundreds, headed for thousands.
3. The consequential work belongs to engineers. Problem-solvers who can reason through ambiguity, distributed systems, and integration risk become the scarcest and most valuable talent.

The architectural bets:

4. Machine-scale infrastructure. Git itself being reengineered, the monolith giving way to API-first composable services, agent-specific APIs as first-class.
5. Orchestration across the full lifecycle. CI/CD reimagined as the runtime that coordinates agents, validates work, and enforces guardrails at machine rate.
6. Context as the superpower. A connected data model spanning planning, code, review, security, deployment, and operations, exposed as a first-class API-accessible service.
7. Governance built into the core. Identity, audit, policy, and deployment flexibility as core platform services, not bolted on.
8. One platform, three modes. Human-owned, agent-assisted, and agent-autonomous work, cloud and model neutral.

How they will deliver:

9. A flexible business model. Subscriptions plus consumption pricing, with more mixing as work evolves.
10. A culture of excellence. Three new operating principles: Speed with Quality, Ownership Mindset, Customer Outcomes.

That is the whole frame. Let me say what I think it means.

This Is Empowered Teams at $1B+ ARR

The single most important sentence in the entire letter, for me, is buried in the operational moves: “re-organizing R&D to create roughly 60 smaller, more empowered teams with end-to-end ownership, nearly doubling the number of independent teams.”

If you have read my earlier piece on leading AI as an empowerment problem, you will see this immediately. Marty Cagan has been arguing for two decades that the size and shape of the team is the first lever a product organisation pulls. Empowered teams - small, durable, cross-functional, owning a problem rather than a feature backlog - are how good product companies move fast at scale. EMPOWERED the book has been on the shelf of every reasonable CPO since 2020.

What GitLab just did is the operational reality behind that bookshelf. Nearly doubling the number of independent teams is not a cost play. Cost plays consolidate. Empowerment plays multiply. You only run this move if you believe that the productive unit of the agentic era is the small, autonomous team, not the large, coordinated org. “The people closest to the work make the decisions about it, and they own the result. Layers of management between leaders and the work coming out, and handoffs that dilute accountability are eliminated.” That is Staples writing the same line Cagan and Hastings have written, but now as the operational principle of a public company.

The flattening - removing up to three management layers - is the dual move. You cannot run sixty empowered teams through eight management layers. The math doesn’t work and the latency doesn’t survive. If you want context-not-control, you have to delete the control surface, not just decline to use it. That is what flatter org means in practice.

I will be watching this carefully because it is the highest-profile validation yet of the empowered-team thesis in the AI era. If it works, every other dev tools, infra, and SaaS CEO is going to be on a board call within a quarter explaining why they aren’t doing the same thing.

Software Will Be Built by Machines, Directed by People

Belief 1 of the ten is the load-bearing one: “Software will be built by machines, directed by people.” This is exactly the thesis I have been working through, just with a more confident verb. Eric Schluntz’s framing in his vibe coding talk was forget the code, not the product. Kief Morris’s framing in the Fowler piece was be on the loop, not in it. Bill Staples’s framing is the executive version of both: the machines do the work, the humans direct it.

The distinction Staples then draws is important and I want to flag it because most enterprise readers will skim past it:

Humans still own the judgment that matters most: architecture, deep understanding of the customer problem, the tradeoffs that require taste.

This is the empowerment frame fully internalised. The human role is not less - it is concentrated at the points of highest leverage: architecture, judgement, customer insight, and taste. This is exactly Cagan’s “work on problems instead of solutions” applied to the agentic era. The engineers and product managers who can do the consequential work become more valuable, not less. The ones who were adding value by being a fast typist or a careful reviewer of obvious diffs will not. That is the talent-density inflection Reed Hastings warned about, arriving on schedule.

The Architectural Bets Are Each a Bigger Bet Than They Look

The five architectural bets in the letter deserve their own consideration. Three of them in particular are bigger than their bullet points suggest.

Machine-scale infrastructure: rebuilding Git

This one is bold to the point of audacity. GitLab is publicly saying Git itself is being reengineered for machine scale. Not Git wrapper services, not Git-on-top, not new APIs around Git. Git itself. The reasoning is sound: agents open MRs in parallel, trigger pipelines around the clock, push commits at machine rate. The 2005 design assumptions about human-rate commits do not survive that load. If you are an enterprise running a hundred agents in parallel, you find this out the hard way the first month.

This is the kind of bet that only makes sense if you believe Belief 1 deeply. If you believe agents are going to be doing most of the building, then the substrate has to be rebuilt to handle that. The competitive moat created by being the platform that handles agent-rate work as the default, while everyone else bolts AI onto a human-shaped Git, is enormous if they pull it off.

Context as a connected data model

Belief 6 is the one I had to read twice. “What doesn’t commoditize is the unique context the model gets to work with: a data model that connects planning, code, review, security, deployment, and operations across every project and repository, accumulated over years of a team’s work.”

This is GitLab realising that they own a longitudinal dataset that no AI lab can replicate. OpenAI can train a better model. Anthropic can train a better model. Neither of them can synthesise five years of your company’s planning-to-production data. The platform that owns the connected graph wins the agent-effectiveness battle even if the underlying models commoditise. This is, in my opinion, the deepest insight in the entire letter. It also explains why every dev tools company that has fragmented planning, code, review, and ops into separate products is suddenly in a worse position than they realise.

One platform, three modes

The third belief that I want to flag is “one platform, three modes” - human-owned, agent-assisted, agent-autonomous. This is the only realistic shape of the next five years. Anyone who tells you all software is going fully autonomous tomorrow is selling you something. Anyone who tells you nothing important will be agent-autonomous in two years is also selling you something. The truth is what every enterprise CTO already knows: you will have a portfolio of modes across your codebase, and the platform that handles all three with the same data model and the same governance wins. It is the same logical structure as hybrid cloud, fifteen years later, applied to the human-agent axis instead of the on-prem-cloud axis.

The Operating Principles Are the Cultural Operationalisation

Staples retired CREDIT, GitLab’s long-running values framework, and replaced it with three new operating principles:

1. Speed with Quality
2. Ownership Mindset
3. Customer Outcomes

Each is unpacked with behavioural expectations. The full lists are in the letter and worth reading, but I want to highlight the line in Speed with Quality that is the entire empowered-team-meets-agentic-era thesis in one bullet:

If an agent can do it, we automate it, and find things where our judgement or skill is essential.

This is the discipline of Morris’s harness, the spirit of Cagan’s empowered teams, and the rule Schluntz proposed - all in a single bullet point that every GitLab employee is now expected to operationalise daily. Find the leaf nodes, automate them, and concentrate human attention on what genuinely needs it. This is the operating principle of the next decade of software companies, articulated as a culture rule rather than an architecture rule.

The Ownership Mindset bullet “It is never someone else’s problem” and the Customer Outcomes bullet “My work creates joy and delight for customers so they love GitLab” are old empowered-team principles, and they read like EMPOWERED and No Rules Rules condensed into team rituals. Good. The principles that matter are the ones that survive across eras. The AI era doesn’t replace them, it sharpens the cost of not living by them.

What I Find Risky

I am not a sycophant about this kind of move and the article isn’t useful if it doesn’t say what I’d push back on. Three risks worth being honest about:

1. The transparency tradeoff

Staples is doing this transparently with a voluntary separation window. That is genuinely admirable and probably the right call ethically, but it carries a real execution cost: several weeks of uncertainty across the workforce while planning is in flight, with the people best positioned to leave being precisely the ones a winning Act 2 needs to keep. The Keeper Test from No Rules Rules is the relevant frame: the high-performers can always go, and a long uncertainty window plus a voluntary package will accelerate departures of people you cannot afford to lose. The honest version of this risk is that the transparent process is expensive in talent density, and the bet is that the gains in trust outweigh those losses. I think Staples is probably right but it is not free.

2. Rebuilding Git while customers depend on you

Belief 4 commits to a generational rebuild of the underlying infrastructure - Git itself, the monolith, agent-first APIs - “without disruption to GitLab customers that depend on us every day.” This is the platform-engineering version of changing the engines while flying. Every CPTO who has shipped a generational infra rebuild knows the failure mode: timelines slip, the old surface keeps consuming engineering capacity, and the new surface ships eighteen months late with feature gaps that customers were promised. The fact that they explicitly call out no disruption is right but it is the hardest sentence in the letter to deliver on.

3. Right-sizing roles before harness is proven

Belief 4 of the operational moves - “rewiring internal processes with AI agents, automating the reviews, approvals, and handoffs to speed us up, and plan to right-size roles across the company to follow suit” - is the move that needs to happen, but the order matters. The hardest lesson of agentic engineering at scale is that the harness has to be built and proven before you remove the humans whose judgement was previously absorbing the harness’s gaps. Right-sizing too early creates the same failure mode I wrote about in vibe coding in prod - debt compounds in places no one is watching because the verification layer hadn’t been built yet. I trust GitLab’s engineering culture to get this right, but it is the place where this whole programme could quietly slip from agentic empowerment into agentic understaffing. The watchword is prove the harness, then resize.

What This Validates

Stepping back, what does this letter validate about everything we have been writing about?

Claim from earlier articles	What GitLab’s Act 2 says
Empowered teams are the productive unit of the agentic era	”Roughly 60 smaller, more empowered teams with end-to-end ownership, nearly doubling the number of independent teams.”
Flat organisations beat layered ones once agents are in the loop	”Up to three layers of management” removed
Context, not control, is the leadership posture	”The people closest to the work make the decisions about it, and they own the result.”
The human bottleneck is at architecture and judgement, not at code review	”Humans still own the judgment that matters most: architecture, deep understanding of the customer problem, the tradeoffs that require taste.”
Verifiability and the harness, not inspection, are the way to manage agents at scale	Orchestration as runtime, governance built into the core
Connected, longitudinal context is the real moat	”Context is what lets agents spend fewer tokens and deliver better results.”
Leaf nodes and trunks - automate where you can, concentrate humans where they matter	”If an agent can do it, we automate it, and find things where our judgement or skill is essential.”

If you had told me six months ago that a public dev-tools CEO was going to ship the cleanest single-document version of every argument I have been making, I would have been polite about it. Staples did it on May 11. The press will cover the workforce reduction. The press will under-cover the strategic thesis, which is the part that matters for the next decade.

What I Am Going to Steal

There are three operational moves in this letter that I am going to steal for the London CPTO role I am about to start:

1. Operating principles, not values. GitLab retiring CREDIT in favour of Speed with Quality, Ownership Mindset, Customer Outcomes is the right shape of move. Values frameworks describe who we are. Operating principles describe how we work. The agentic era rewards the second framing, because “if an agent can do it, we automate it” is an operational rule that you can act on in a daily standup. “We are open” is a value that requires interpretation. I want operating principles, scoped tight, with concrete behaviours, that we can hold ourselves to.
2. Right-sized empowered teams, day one. I am not going to scale into empowered teams. I am going to start there. Small, durable, cross-functional, owning a problem, with the harness built up around them. If we ever feel the need to add a layer of management, that is the moment to ask why the harness isn’t doing the job.
3. Customer zero, agentic. GitLab’s phrase “customer zero” for using their own platform on themselves is exactly the right discipline for AI tooling. If we cannot vibe code in prod responsibly on our own product, we have no business selling it as the answer to anyone else.

The Takeaway

The boring, conventional read of GitLab’s Act 2 is: another tech company restructured because of AI. The accurate read is: the first public company has openly bet its operating model on the empowered-team-meets-agentic-era thesis. The structural moves - flatter, smaller, empowered, agent-rewired - are the most concrete real-world implementation we have seen of what the next decade of engineering organisations will look like.

Whether or not GitLab gets it right operationally, the playbook is now public. It will be borrowed, adapted, and re-applied across the industry within two quarters. Every CPTO who has been waiting for proof that the empowered-team playbook still works in the AI era now has the answer they were waiting for: yes, and the public companies have started.

If you read one CEO letter this quarter, read this one. If you read two, pair it with Hastings’s No Rules Rules and Cagan’s EMPOWERED. The argument is the same. The execution is now public.

References

• GitLab Act 2 - Bill Staples - The CEO letter to customers, investors, and the team
• Leading AI Is an Empowerment Problem - The framing this article validates
• Vibe Coding in Prod Is a Management Problem - The individual-contributor companion
• EMPOWERED - Marty Cagan (book notes) - The empowered-team playbook
• No Rules Rules - Reed Hastings (book review) - High-talent-density, candor, context not control
• Three-Tier AI Architecture - When agents fit where
• The Last Bottleneck - The exponential and the human bottleneck
• The Skills Framework - The procedural discipline layer
• Agentic Development Patterns - The daily workflow
• The Overnight Agent Factory - The compounding mechanism