The Triad Coordination Tax
Engineering is collapsing into product. So is design, so is middle management. The coordination tax between them is shrinking.
In a recent client engagement, I’ve been working with investment analysts. A small tech team works alongside the business. The analysts themselves are building their own POCs — not through a PM, not through a backlog, not through a request-for-proposal that turns into a Jira epic six weeks later. They prototype directly. And they do it with Claude Code. The small tech team (us) picks up where the business’s prototype runs out of road, expands it, and productizes what survives.
Five years ago, this engagement would have been a delivery team of six+ on the vendor side — PM, designer, frontend, backend, devops, you name it — handing off to a passive business sponsor on the client side.
So, what happened?
The seats that have disappeared are the seats that existed to translate between the business and the engineering.
That is the article in one anecdote. What used to flow through a chain of specialists — each translating the work for the next — now flows directly. The seats that survive on both sides are smaller, more senior, and more shaped.
In the previous essay, I described what I think is happening to the composition of the software engineering workforce. This is the companion question:
What shape does the surviving organization actually take?
I believe there are three threads to this.
The Triad was a coordination structure
For about fifteen years, the canonical tech org had three sub-organizations. Engineering. Product. Design. The “EPD triad” — popularised by Marty Cagan / SVPG (Inspired, 2008):
In modern teams, we tackle these risks prior to deciding to build anything. These risks include value risk (whether customers will buy it) [Product], usability risk (whether users can figure out how to use it) [Design], feasibility risk (whether our engineers can build what we need with the time, skills, and technology we have) [Engineering], and business viability risk (whether this solution also works for the various aspects of our business—sales, marketing, finance, legal, etc.) [Product].
Inspired: How to Create Tech Products Customers Love (Chapter 7)
by Marty Cagan / SVPG (annotations mine)
It roughly maps to a “three risks” frame:
Product = value/viability
Design = usability
Engineering = feasibility
The triad existed because specialists couldn’t easily understand each other’s work. PMs translated business intent into requirements engineers could execute. Designers translated user needs into visuals PMs could approve. Engineers translated requirements into a working system, and then translated the system’s constraints back into something PMs could communicate to the business.
Cagan’s work became part of the intellectual foundation for product people. By the mid-2010s, it was an industry-wide convention, and organizational structures grew to mimic the frame.
Instead of tight collaboration, we drew organizational boundaries along these dividing lines. (This might have been a distortion of Cagan’s thinking, by the way.)
Every translation now carried a tax — a meeting, a Jira ticket, a Figma handoff, a spec doc, a Slack thread, a manager whose entire job was to enforce the protocol. The tax was the price of having specialists who couldn’t do each other’s jobs.
The cracks start to appear in the facade of this tax, though.
Sahil Lavingia, Gumroad’s CEO, predicted it in a walkthrough of his own workflow back in January 2025. I highly recommend watching it — it has aged remarkably well, in times when material from last month is often already outdated. His estimate of the speed gain, in certain cases, is around 40x — a number that sounds absurd until you notice it is mostly the translation time between specialist teams that has been removed, not the time spent doing the underlying work.
A PM who prototypes in Cursor no longer needs an entire engineering team to validate feasibility — the prototype itself is the validation. Sure, it might need proper guardrails and scaffolds, and maybe someone to productionize the output. But that is very different from throwing a ticket over the wall.
Likewise, a designer who can ship their own React doesn’t wait in a queue for engineers (though they might need a code review).
Shopify now includes vibe coding in PM interviews. Think about that. Oh, and Shopify is not the only one. Google, Netflix, and Stripe followed. There is an expectation that a PM can ship a working prototype in 2026.
Maryam Ashoori, leading WatsonX product at IBM, mentions that the traditional 1 PM to 8 engineers ratio is being replaced by 1 PM to 0.5 engineers.
Heck, Marty Cagan himself predicted last year that within 3–10 years we should expect product teams to compress to 3 people.
Joel Lewenstein, the Chief Design Officer at Anthropic, calls it a Venn diagram coming closer together — heavy role collapse upstream in ideation and prototyping, with clearer swim lanes still re-emerging downstream at scale (this is productionization).
The triad is shrinking because the protocol between the three organizations was the expensive part. And a large chunk of that protocol got eaten by AI.
Layers go with it
Some management layers existed to manage the translation. When the translation goes away, these layers have nothing left to manage.
GitLab announced Act 2 on May 11. “Eight layers is too deep for a company our size, and management layers are slowing us down.” They are removing up to three layers in some functions and reorganizing R&D into ~60 smaller teams with end-to-end ownership — nearly doubling the count of independent units. CEO Bill Staples explicitly framed it as “rewiring internal processes with AI agents, automating reviews, approvals and handoffs.”
Meta’s Applied AI engineering team is now running at a 50:1 employee-to-manager ratio. The industry “outer limit” for span of control is 25:1. The research-determined optimum, depending on whose textbook you use, sits around 7. Jensen Huang is notorious for running Nvidia with 60 direct reports and no one-on-ones. But this sentiment is spreading across the industry. Andy Jassy mandated a 15% increase in the IC-to-manager ratio at Amazon at the end of 2024 and doubled down on it in his 2026 shareholder letter. Microsoft cut 7,000 middle managers in May 2025 — without even using the word “AI” in the announcement.
There is a fresh quote from Brian Chesky noticing the same thing: “I don’t think people managers will have any value in the future. When I mean people managers, I mean people that only manage people. (…) You can’t just be these managers where you’re people’s therapists, doing meetings, doing one-on-ones.” The direction is the “manager IC” — every manager has to manage through the work, not around it.
The pure-coordination role — the manager whose job was to be a load-balancer or translator between specialists — is being deleted at the same time the specialists themselves are being merged.
The shaped generalist comes back
If specialist teams merge and translator managers go, what shape is left? There’s a third thread. The people who survive look different from the people who got the company here.
The 2010–2022 era of hyper-specialisation — backend engineer, frontend engineer, infra engineer, ML engineer, data engineer, mobile engineer, QA engineer, SRE, DevEx, four flavours of designer, three flavours of PM — was a function of two things. Cheap capital, which let companies hire one of everything. And the limits of any one human’s ability to hold the whole stack in their head.
Both of those constraints have softened. Capital is no longer free. And the assistant you sit next to can hold whatever part of the stack you forgot.
What comes back is the T-shaped generalist. An engineer who can write the spec, ship the prototype, validate with users, instrument the metric, and own the operational tail. A product person who can prototype the thing they used to brief. A designer who ships their own UI to production. The early-stage founder profile, basically — but as the default shape of a senior contributor at scale, not just at seed.
YC’s Requests for Startups for 2026 names this explicitly: “For the first time a 2–3 person team can ship something a Fortune 10 can find useful before the ink is dry on their incorporation docs.” Tobi Lütke’s Shopify memo made AI usage a basic expectation, and made managers prove AI couldn’t do the job before approving headcount.
Small teams of shaped generalists, with no coordination tax between them, can now do work that used to require the organizational triad plus three layers of management above it.
Where this breaks
It’s worth being honest about where the collapse stops.
Late-stage product work still has swim lanes. Lewenstein’s nuance is the right one — collapse upstream, clearer specialisation downstream as you scale and harden. So it works up to a point. Still, the feasible range is increasing.
Regulated and safety-critical domains — financial services, healthcare, defence, anything with an audit perimeter — still reward people who own a narrow column deeply.
And the span-of-control numbers at the extreme end are going to break some managers. One Wall Street Journal-quoted manager with 36 direct reports described a “constant state of inadequacy”. I can tell you I’ve already had multiple conversations like that. No one said the transition was going to be easy.
And yet, my take is that removing the layers without redefining how the role gets spread out will produce false efficiency. We’re going to find out soon enough.
The optimistic version of this article is that coordination between the organizational triad was a tax, and the tax is being repealed. The honest version adds: that doesn’t mean the underlying functions disappear. It means they get redistributed across fewer, more capable people — and the people who used to do them as full-time specialist roles need to become something else.
What now?
Play It Again, Sam described the substitution happening at the engineering level — routine work absorbed into a tooled layer, strategic work concentrated and repriced upward.
This is the same mechanism viewed from inside the org chart.
The Triad coordination was a tax.
The layer of managers above the triad was a tax.
The translation between roles was a tax.
What survives is smaller, flatter, and more shaped.
The function persists. The tools change. The people who change with the tools keep their seat.