The AI-Native Leader · Four-part series
What actually changes for engineering leaders when the organization becomes AI-native — and what to invest in before the rubric catches up.
The question isn't whether to be an IC or a manager. It's whether you can design the system the work runs on.
Two questions keep showing up in coaching conversations. Strong ICs asking whether moving into management is still worth it. Experienced managers asking whether they should go back to IC. Different situations, same anxiety — and both are pointing at the wrong problem.
The consensus that companies will need fewer managers isn't wrong. What's wrong is the conclusion people draw from it. Good management was never about headcount coordination; it was about setting context, defining what good looks like, and deciding what gets built. AI didn't eliminate that work. It removed the coordination overhead that was hiding it — which means the part that was always the actual job is now the whole job.
What AI changed is the ratio. Less time producing, far more time deciding. That shift rewards people who think in systems rather than workflows: who can look at how work moves through an organization and redesign the path, instead of optimizing their own step in it.
Nobody has an AI adoption playbook that works, because a playbook is the wrong artifact. What you need is a harness.
Every leadership forum has the same recurring thread: can anyone share an AI adoption playbook that worked? Some of that is pressure from executives who want an efficiency story yesterday. Some is the quieter anxiety of watching everyone else move. Either way, the reach for a playbook is the tell — a playbook assumes the pattern is known.
A harness is the alternative: the scaffolding around the work that decides what gets delegated to agents, what gets judged by humans, and where the handoffs sit. Two harnesses actually, one around the agents and one around the people. Building them is craft, and most leaders already have the underlying skill — it's the same skill as scoping work, setting a quality bar, and giving feedback.
| Mode | When it applies | Example |
|---|---|---|
| Human leads, AI assists | The work is novel; judgment is ongoing. A person decides what questions matter and what good looks like. | Choosing what to build next; org and role design; a feasibility prototype |
| AI drafts, human judges | The pattern is clearer, but quality still needs human eyes. AI does the first pass; a human evaluates what's missing. | A performance review draft; a launch comms plan; a market analysis |
| AI executes, human spot-checks | The pattern is proven and the quality bar is clear. A human watches for drift. | Code for well-understood patterned work; operational review prep |
| AI proposes, human approves | Narrower than it sounds. Not open-ended judgment — work where the consequences of shipping it wrong require a human "go." | Mass customer email; large refunds; regulatory-adjacent submissions |
The real work isn't picking a mode once. It's migration — moving work from mode one toward mode three as the pattern becomes known, and noticing when something has drifted into a mode it doesn't belong in.
The conventional org has many layers. The AI-native one has two patterns of work — and neither is gated by a title.
A conventional R&D org has many function ladders, each many layers deep, connected by cross-functional review forums. The AI-native version collapses most of it. Influence and access to the decision-making room stop being gated by a management title and go instead to whoever runs the harness well, owns a problem end to end, or pushes into the frontier.
Iterative work is the existing business — pricing, onboarding, retention, shipping. Here, one person paired with an agent team drives a piece of work end to end, rather than handing off between roles. Every IC ends up managing a small agent team whether the title says so or not, doing what is unmistakably management: delegation, scope, review, feedback.
Frontier work is the bets that decide whether the company gets a second chapter. It doesn't specialize by role, because the questions are too unstructured for "I'll do design and you do engineering." It pulls generalists — whoever is nearest the question.
Two leader moves matter most inside this. First, bottleneck management: when agents produce five times more drafts, review layers built for human throughput become the choke point almost immediately. Second, end-to-end ownership: let one person own a problem all the way through, and if it's too big, split the problem rather than splitting it across two people.
The work has already changed. The promotion rubric hasn't. Waiting for it to catch up is the expensive choice.
Every company's leveling guide, interview loop, and promotion rubric was written for a version of the work that is already receding. That gap is the thing to manage. If you invest only in what the current rubric rewards, you're optimizing for a scoreboard that's being rewritten while you play.
Three things shift: how you see the work (from producing output to designing the system that produces it), how you see your time (from throughput to judgment density), and how you see yourself (from a specialist defending a lane to someone who can pick up an unfamiliar function through agents).
Some of what you already have folds directly into the new shape. Some overlaps partially. Some is genuinely fresh and has to be built. Sorting your own skills into those three buckets — honestly — is the most useful hour you can spend right now.
It's the wrong axis. In an AI-native org, influence and access to decisions aren't gated by a management title — they go to whoever runs the agent harness well, owns a problem end to end, or is pushing at the frontier. ICs in these orgs already do management work daily: scoping, delegating to agents, reviewing output, deciding what ships. Choose based on which kind of work you want, not which chair it comes with.
Most likely yes, in raw count — but that's not the useful conclusion. What AI removed is coordination overhead, which was never the real job. What's left is the part that always mattered: setting context, defining what good looks like, deciding what gets built. That work grew, it didn't shrink. Fewer managers, doing more of the actual job.
It sits at the far end of a spectrum. AI-augmented means tools bolted onto processes designed for how people work. AI-redesigned means process and organization rebuilt around what AI can actually do, with people still at the center of execution — roughly where the AI labs operate. AI-native means the core of how the organization runs is built on AI, with people focused on bringing in outside signal and steering the system.
Because adoption was the easy part, and it's the part that shows up on a dashboard. The two things that actually block the step change are process built for human coordination — reviews, approvals, alignment meetings that become the bottleneck the moment agents produce more drafts — and organizational knowledge trapped in individuals' heads, which AI can't reach, so its output stays generic.
System design over personal throughput; judgment density over output volume; and the ability to pick up an unfamiliar function through agents rather than defending a specialty. Invest ahead of your company's promotion rubric — it was written for a version of the work that's already receding.
I coach tech leaders navigating exactly this shift — what to build, what to let go of, and how to invest ahead of the rubric.
Book a free clarity call Subscribe free