2026-05-31
May 31, 2026
Issue 022 — The org chart is shrinking under you. The unit of leadership status — "I run a team of N" — is depreciating fast. The leaders who get the next decade right are the ones who already swapped scoreboards.
A friend who runs platform engineering at a large fintech told me last week he was being "promoted" — same title, same scope on paper, half the org. The middle layer he'd built over four years had been removed in a single restructuring. His skip-levels now reported to him directly. The headcount line on his packet went down. The decisions-per-week line went up by maybe 4x.
He asked me, half-joking, how to spin this to his peer group. I told him to stop spinning and start telling the truth: the thing he used to be measured on is no longer the thing that matters, and pretending otherwise will get him outflanked by leaders half his level who have figured this out.
This is not a fintech story. ICONIQ Growth's latest GTM report shows top-quartile public software companies running with management ratios 9x flatter than the median — not as an emergency cost-cut but as a deliberate design choice. SaaStr is calling $500K ARR per employee "the new $200K." Marketplace ran a segment ten days ago naming the trend the "Great Flattening." The middle, where most TPM and tech-leader scope lived, is the thing being removed. And nobody has rewritten the scoreboard.
For roughly thirty years, the unit of organizational status in tech has been people. How many engineers report up to you. How many skip-levels. How many open headcount reqs you control. Promotion packets, comp bands, and the social hierarchy of an offsite all encode the same currency: you are the size of your org.
That currency made sense in a world where the binding constraint on shipping software was human throughput. More people meant more code, more code meant more product, more product meant more revenue. Sponsorship and political leverage followed headcount because headcount produced the output.
In 2026, the binding constraint isn't human throughput anymore. Coding agents have absorbed enough of the implementation surface area that the median feature ships with measurably fewer engineer-hours than it did eighteen months ago. ICONIQ's data shows companies with AI fully embedded in their workflow generating roughly 2x net new revenue per FTE versus low adopters. The flatter, leaner org isn't a downturn artifact — it's the high-performing configuration. The market is rewarding it, and operators are designing toward it on purpose.
When the binding constraint shifts from throughput to direction, the scoreboard has to shift too. The thing that's actually scarce in a flat, AI-leveraged org is good decisions made fast. Not code. Not headcount. Decisions — about what to build, what to kill, which bet to double on, which hill to die on, who to hire one of, what to standardize, what to leave intentionally fragmented.
Will Larson, now CTO at Carta, has been pushing this point in Crafting Engineering Strategy: most failed engineering orgs don't fail at execution, they fail at the small number of strategy decisions that should have been made earlier, sharper, and with more conviction. The senior leader's job is not to do more — it's to decide more, faster, and at higher quality than the org could on its own.
If decisions are the unit, three things follow that most senior leaders haven't internalized.
One: the size of your org is no longer a reliable signal of your leverage. A director with 80 reports who runs three real decisions a quarter is doing less leadership work than a staff engineer with no reports who makes thirty good calls a week on architecture, prioritization, and what not to build. The org chart still shows the director as bigger. The output and the comp will eventually disagree. Comp is already starting to disagree at the top of the IC ladder — Charity Majors has been right for nearly a decade that the engineer/manager pendulum would become a default rather than an exception, and the flattening is what tips it from preference to structure.
Two: most of your leadership debt is decision debt. Open decisions you've been carrying for weeks because the data isn't perfect, or the stakeholders haven't aligned, or you're waiting for the right meeting. Each of those is now a binding constraint on a leaner org that needed the decision two sprints ago. Annie Duke's frame on this is the right one: decision velocity isn't recklessness; it's calibration. For most decisions, the cost of waiting exceeds the cost of being wrong, because most decisions are reversible. The leaders I trust most have stopped optimizing for being right and started optimizing for deciding fast, watching for signal, and reversing without ego.
Three: your job is to multiply the org's decision rate, not to be the bottleneck. The flat org is not a flat decision graph — it's a flat reporting graph that runs only if decision authority is genuinely pushed down. The senior leader's craft is now (a) decide the small number of one-way doors yourself, with care, (b) push two-way doors down to whoever can decide soonest, with a budget for being wrong, and (c) build the artifacts — strategy documents, written principles, "here's how we decide" memos — that let the org make 80% of routine decisions without you in the room. Without (c), pushing decisions down becomes "you decide, I second-guess," which is worse than centralizing.
If you want a concrete test of whether you've made the shift, run this audit for two weeks. Each Friday, write down: how many decisions did I make this week? How many did my org make without me that I would have wanted to make? How many of either category are now actually being executed against? A senior tech leader operating well in 2026 should land somewhere north of 25 personally-made decisions a week, with another 50–100 being made under written principles they wrote. If those numbers are in the single digits and your headcount is in triple digits, the org is leveraged for the old scoreboard. It will look strong on the org chart and feel slow in the market. That gap is what the flattening is correcting for, whether you participate in the correction or not.
The uncomfortable part of this shift is that it forces senior leaders to put their judgment on the table much more frequently and much more visibly than the headcount era did. You can hide behind a big org; you cannot hide behind a public decision log. That's a feature, not a bug. The leaders who'll matter in 2030 are the ones who are voluntarily moving into the visible-judgment regime now, while everyone else is still polishing the org chart.
Try this week. Pick one decision you've been carrying for more than 10 days. Decide whether it's a one-way door or a two-way door. If two-way: decide it today, in writing, with a 30-day review. If one-way: schedule a 90-minute block this week with the two or three people whose dissent you most need, and either close it or write down exactly what evidence would close it. Add the outcome to a running decision log you share with your skip-levels. The log itself is the artifact — start it.
What it is. A two-category sort that asks one question of every pending decision: if this turns out to be wrong, how reversible is it? Two-way doors are reversible — you walk back through, mostly intact. One-way doors are not — you live with the consequences for a long time. The rule: decide two-way doors fast, with whoever is closest to the work; reserve heavyweight, slow, senior-led process for the one-way doors only.
When to use it. When you find yourself spending equal calendar time on a hiring decision and a quarter-defining strategy decision. When meetings keep re-litigating the same call. When a leaner org is making you the bottleneck on routine choices you don't actually need to own. When you suspect your team is treating reversible bets like irreversible ones (the common failure mode) — or, less common but more dangerous, irreversible bets like reversible ones.
How to run it:
When NOT to use it. When the decision is genuinely small enough that even the sort is overhead — let it be made and move on. And when the org is in acute crisis: a real incident needs OODA, not door-classification. The framework is for the steady-state portfolio, not for the burning building.
Example: a senior platform lead I work with ran this audit on a 14-item open-decisions list. Eleven were two-way doors that had been waiting on her for an average of 19 days. She decided ten of them in a single 90-minute block, delegated the eleventh, and put a 30-day review on each. Two months later: seven held, three were reversed cheaply, none of the reversals cost meaningful money. The org's decision throughput roughly tripled, and her own calendar opened up enough to actually work the three real one-way doors that were buried underneath.
The Modern GTM Org in 2026: 9x Flatter, 20–30% Leaner — ICONIQ's latest cut shows top-quartile orgs deliberately running with 9x flatter management ratios as a design choice, not a downturn artifact. The same logic is moving into R&D. If your org is still defaulting to the median structure, you're optimizing for a benchmark that's two years stale.
Call it the "Great Flattening" — Marketplace's framing of the middle-management cull names the trend cleanly. Worth ten minutes for the language alone — naming a trend is half the work of getting your exec team aligned on responding to it.
Engineering Leadership in an AI Era — ICONIQ's companion piece on the R&D side. Notable data point: AI-mature engineering orgs aren't just shipping more, they're shipping with materially fewer managers per IC. The "fewer managers" part is where the leadership career-pathing conversation has to start.
"The quality of a leader's decisions, divided by the time it took to make them, is the most important metric in a flat organization. We've been measuring the numerator and ignoring the denominator for twenty years."
— paraphrased from a senior CTO peer group conversation, May 2026
Don't miss what's next. Subscribe to Critical Path: