Six months into my job, the PE firm that owns the company asked me to help its portfolio companies find people who run engineering organizations and still build things themselves. Leaders, up to their elbows in the work.
Two years ago, I’m not sure I could have told them where to look. I probably would have recommended against it. It ran counter to my management philosophy: “you’re not being paid to do things, you’re being paid to lead.” As head of engineering for Tripadvisor, I heard an enduring story from my coach about the “do nothing” manager, someone who expressed his value entirely through delegation. Not a bad lesson...back then.
When this group first met in my living room in June 2025, nine people fit the description, including about half of those who came back this year. Not one of us had a name for it. I wrote that afternoon up as The Convergent Mind, published June 25, 2025. Read back now, it is a careful catalog of what those people believed and a blank space where the thing they had in common should have been.
We convened a similar room on July 24, 2026. There were ten people, and six had built something to show. In twelve months the role acquired at least six competing names.
TL;DR
A year ago the hands-on leader was rare enough that a room full of them had no shared word for it. Now the role carries at least six circulating labels, and the PE firm that owns the company I work for has asked me to source people who fit it.
LeadDev’s survey program reversed itself inside twelve months. Its June 2025 reporting had hands-on opportunities for leaders drying up under budget pressure. Its 2026 report (n=600) has 37% of leaders doing more hands-on technical work and 33% of managers considering a move back to individual contributor.
That same 2026 report also has 45% working more hours and 41% reporting team motivation down.
The scale of what a single practitioner builds changed categorically. In 2025 the frontier in this room was prompt scaffolding and a CRM made of Markdown files. In 2026 it was token-cost governance with FinOps conversations at CFO level, agent monitoring as its own sub-discipline, and personal systems resolving years of accumulated data behind an agent interface.
A counter-current showed up that the 2025 conversation had no room for. One founder in the room thinks the advantage any of us build individually is temporary, because the frontier labs will absorb it.
What a Leader-Practitioner Is, and Why AI Made It Possible
The role is easy to describe. Someone who runs an organization and does hands-on technical work as a deliberate part of the job, not in the gaps between meetings. The work has a recognizable shape: internal tooling, agentic harnesses, MCP connectors, a DX platform layer other teams depend on. Work that enables production rather than sitting on its critical path. At startups, that can extend to production code. I know more than one where the CTO writes all of it.
What changed is the time-and-cost proposition. Directing a team of agents is close enough to delegation that it fits inside a leadership mindset, which sustained hand-coding rarely did. Now you specify intent, review what comes back, correct, and direct again. A fifteen-minute attention window can produce something that used to need an uninterrupted day.
I made the full case for the archetype, where it comes from, and why it was rarely adopted for most of the last forty years in The Era of the Leader/Practitioner, published June 1, 2026. This piece is about what happened to the role in the past year.
A Year Ago This Was Rare. Now Companies Recruit for It.
The signal I trust most is that capital is staffing for it. A private equity firm asked me to help find these people for companies it owns, and a startup I advise already has one. I’m informally helping a few others. A year ago, I had not seen anyone recruit against this description because the job did not exist in a form you could put in a posting.
The survey data moved the same way, and fast enough to cause whiplash. LeadDev’s June 2025 reporting described hands-on opportunities for engineering leaders as having dried up. Its 2026 Engineering Leadership Report (n=600) has 37% of leaders doing more hands-on technical work and 33% of managers considering leaving management for an IC track. One survey program, one year apart, and the message reversed.
The same 2026 report has 45% working more hours and 41% reporting team motivation down. It records leaders doing more technical work inside organizations that are, by their own account, more tired.
What has not happened is agreement on what to call any of it. Gergely Orosz predicted the player-coach in 2024, and the term is heavily used in LeadDev’s coverage as recently as July 27, 2026. Addy Osmani introduced loop engineering at O’Reilly Radar on June 22, 2026. “Agent orchestrator” circulates widely, including one vendor calling it the most important job of 2026. “Founder mode” has been around since Paul Graham wrote up Brian Chesky in September 2024. Simon Willison offered “vibe engineering” in 2025 as the serious counterpart to Karpathy’s vibe coder. Mine is leader-practitioner, which makes six labels for one job. James Stanier’s “The End of the Non-Technical Engineering Manager”, for LeadDev on April 20, 2026, skips the naming problem and describes what the role displaces.
The description cuts across titles. I opened the introductions by calling one person there the most emblematic of the era we were in. His reply:
“I’m 61 years old and this is the first time I’ve been emblematic of an era.”
He was not an engineer at all, and he stayed on the same note when he introduced himself:
“Everybody remember Coco, the gorilla who learned sign language? I think I’m sort of that in this room.”
That was a longtime marketing consultant who shipped his first piece of software, ever, this year. Its market fits in one sentence:
“My market is people who have one to three short-term rental properties that they self-manage and are scared shitless of losing money.”
An angel investor who had come up through IT consulting put the same impulse differently:
“Having been a technologist for most of my career, I can’t let new tech get away from me.”
And then:
“I think probably there’s at least 50,000 other people building the exact same thing.”
The 2025 piece did not anticipate this. It described nine people who had converged on nearly identical conclusions about how software gets built. It closed on new patterns and organizations. It had a great deal to say about what we thought and little about what we were.
The Scale Changed
The group changed more than its headcount. What one person now runs would not have been buildable twelve months earlier, and the problems people brought to the room only exist at the new scale.
An advisor who works with engineering leaders on AI adoption had turned the agents in his personal system into a named executive team:
“Sloan’s my head of engineering. Louise is my editor in chief. Morgan is my chief operating officer. Ren is my head of design.”
He also had a cross-lab failover rule. If Anthropic went down, one switch would move the system to another lab’s models within minutes.
The head of engineering at a research firm demoed a token dashboard. It started with a bill nobody could explain:
“So, yes — this is something I built in about nine hours. Why was it built? We were going through a token-maxing issue — a really bad token-maxing issue, where nobody knew how much they were spending.”
Six months earlier, the provider APIs behind that dashboard did not exist.
A Brooklyn-based startup operator and fractional CTO had arrived at monitoring from the other end, watching his own agents rather than his own spend:
“I have a Hermes agent that runs. I’m always trying to throw more work at it. So I kind of built a tool to help keep tabs on what it’s doing.”
His tool is deterministic rather than agentic, runs roughly a hundred monitors, and feeds a self-remediation loop back into the agent it watches. He introduced it with no ceremony whatsoever:
“So this is called, I call this ‘Glance’.”
Both people built instrumentation for processes they no longer needed to watch directly. Those processes existed because of the AI tools they had added. No one in the 2025 group brought a problem like that.
The angel investor did not demo, but the people who presented were not the only builders. He described a personal system that pulls his accumulated records into a single graph, resolves the same person across the addresses and jobs they have moved through, and sits behind an agent interface he can ask to brief him before a meeting. In Is Your Digital Brain the Light Saber?, I argued that building your own knowledge system rather than renting a generic one separates practitioners now. He built one, though he does not write software for a living.
Technical terms have become vernacular. The marketing consultant, describing what he liked in a recent model release, reached for a term the room’s engineers use as jargon:
“The thing that I really liked about Fable is that it introduced the idea of the adversarial review.”
A serial founder supplied a counter-current the 2025 conversation had no room for. He had built something and concluded that the market for it was disappearing:
“Two weeks — less than two weeks into that, I realized shit, the problem I solved is going away. It’s financial.”
“The combination of mixture-of-experts models and optimized model routing are probably going to eat a lot of that.”
The 2025 piece framed convergence as validation. This group took for granted that the value of individual construction is shrinking while the volume of it grows.
Software generation rose even as its durability decreased. Ten people who fit a description that barely existed a year ago spent an afternoon comparing instrumentation for systems they did not need to observe back then. My own engineering organization, meanwhile, still operates much as it did last year. The tools are everywhere: JetBrains’ August 2026 follow-up, based on more than 15,000 professional developers surveyed from May through July, found 90% using AI coding agents at work at least weekly and 68% using them daily. Most do not work like we do. Part 2 is about that gap.
Bob Matsuoka is CTO of Duetto, a hospitality profit and revenue-management platform, and writes about AI-augmented engineering practice. Previously, Bob has been CTO of Tripadvisor, Citymaps, and Runtime Technologies.
As in last year’s piece, the people in this room are identified by role rather than by name. That was the condition of the conversation, and it holds here.
Related reading:
The Convergent Mind — The same gathering, one year earlier, and the piece this one grades.
The Era of the Leader/Practitioner — The full case for the role, including where the archetype comes from.
AI Power Ranking — Tool comparisons and benchmarks for AI practitioners.
LinkedIn Newsletter — Strategic AI insights for CTOs and engineering leaders.
This is fascinating, Robert, particularly because I wonder whether AI is changing the leader/practitioner role OR is laying bare something already wrong with how organisations have frustratingly separated expertise and leadership. Or perhaps even in society.
Our organisations implicitly tell technical experts like engineers that career progression means moving further away from the work that created their credibility in the first place. You have to be a leader or an expert. You can't be both. So choose now.
But that's a bugbear of mine that I will take every opportunity to raise that I can. However...
I wouldn't necessarily conclude that the answer is for leaders to start doing more of the technical work again.
The question I'd be asking is: "Where does their expertise now create the MOST organisational value?"
Sometimes that will mean getting their hands dirty. But as AI progresses, senior expertise will create value via their judgment: framing the right problem, recognising patterns others can't see, challenging assumptions and bias, making better decisions and increasing the overall bench strength of their team and organisation.
AI can dramatically increase someone's capacity to do the work. I'm less certain that it increases their capacity to judge whether that work should be done in the first place. And the more we rely on AI, how do younger engineers develop their judgement?
Perhaps the emerging role isn't simply leader vs practitioner. It's someone who knows when to do, when to lead, and when their expertise is most valuable by enabling somebody/something else.