2026-10-04
October 4, 2026 · Issue 114
Flattening removed the managers who relayed information and AI is automating the relaying itself. What remains is the work that was never relay: deciding what is true, what is next, and who owns it.
For a decade, a large share of technical leadership, TPMs included, was connective tissue. You carried context between teams, translated between levels, assembled the status, and chased the dependency. It was real work and it was valued, mostly because information moved slowly and someone had to move it.
Two forces are now compressing that work from both sides. Organizations are flattening, and AI is absorbing the coordination tasks that remain. James Stanier makes the case in LeadDev that managers whose value rested on being "connective tissue between people and teams" face the largest career risk, and he cites a Harvard Business School study finding that after Copilot's introduction coding activity rose 12% while project management activity fell 25%. That is one study, so hold it loosely, but the direction matches what most of us see in our own calendars.
The comfortable reading is that this is a threat to other people's jobs. The more useful reading is that it is a reallocation of what "senior" means. The essay below argues that the durable roles are judgment roles, and that you can build the muscle deliberately.
Start with the evidence that the role is already changing shape. The LeadDev Engineering Leadership Report 2026, a survey of 600 engineering leaders published in September, found that 37% are doing more hands-on technical work, 45% are working longer hours than a year ago, 41% say their teams are less motivated, and one in three managers is considering leaving management for an individual contributor role. Stephane Moreau argues in LeadDev that the engineering manager role is splitting in two: a Tech Lead Manager running three or four people while still writing code, and a multi-team manager with shallow technical depth per team. He notes Meta's AI team reportedly runs at around fifty reports per manager, against the traditional seven.
Deloitte's 2026 Global Technology Leadership Study frames the same shift from the top. Seventy-nine percent of tech leaders say they now prioritize business outcomes over operational management, 71% of organizations have five or more technology leaders, and 42% report minimal or zero ROI on AI investments. Read those together: more leaders, less command authority, and a large gap between AI spend and AI return. Someone has to adjudicate that gap. That someone is not a status aggregator.
Here is the claim. When coordination gets cheap, judgment becomes the scarce input. Will Larson's revised rules of engineering leadership points the same way: migrations that used to take a team can be owned by one engineer, and the leverage moves to the development harness (tests, CI, validation environments) and to humans reserved for edge cases and high-value calls. Generating the work is cheap. Knowing whether it is right, safe, and worth doing is not.
I would break the judgment layer into four jobs, and each is a place a TPM or tech leader can stake a claim:
1. Framing. Deciding which problem the organization is actually solving, and whether it is complicated or complex. Agents will happily execute a plan for the wrong problem at ten times the speed.
2. Sequencing under uncertainty. Choosing what goes first when dependencies are real and information is partial. This is the old critical-path skill, but the inputs are now fuzzier because the cost of building is falling faster than the cost of integrating, verifying, and operating.
3. Verification design. Deciding what evidence counts as "done" and building the harness that produces it. If review is the bottleneck, whoever designs the review gets the leverage.
4. Accountability architecture. Making ownership explicit when authority is distributed. With a median technology organization now run by five or more leaders, ambiguity about who decides is the most expensive defect.
Notice what is absent: relaying, summarizing, scheduling, reconciling two spreadsheets. Those will be done by tools, and you should be glad.
The risk is that the judgment layer is harder to see. Status reports are legible. Good framing is invisible until it prevents a failed quarter. So the practical advice has two halves. Build the capability: practice making forecasts you can score, write decision memos that name the alternatives you rejected, and keep a log of calls and outcomes. And make it legible: leaders promote what they can see, so show the decision, the options, and the result, not only the output.
One more caution from the same data. A leadership vacuum is forming as middle layers shrink. Kelli Korducki reported in LeadDev that staff engineers are increasingly managing "in everything but name," with little training or pay for it. If you are a TPM, that vacuum is an opening and a trap. Fill it on purpose, with a defined scope and an explicit mandate, or you will absorb it as unpaid, unrecognized coordination, which is exactly the job being automated.
The future is not fewer technical leaders. It is fewer relays and more judges. Pick which one you are building toward.
Try this week. Open your last five status updates. For each, mark every sentence as relay (information someone else already had) or judgment (a call, a risk assessment, a recommendation with a rejected alternative). If judgment is under a third of the content, rewrite the next update so it leads with one decision you need or one call you are making, and cut the rest to links.
What it is. A technique for preparing decisions against several plausible futures rather than one forecast. You build a small set of internally consistent stories about how key uncertainties could resolve, then test your strategy against each.
When to use it. Multi-year platform, staffing, or AI-adoption bets where the outcome depends on forces you do not control, and a single-point forecast would give false comfort.
How to run it:
When NOT to use it. When the decision is near-term and reversible; run a cheap experiment instead.
Example: cross "AI makes build cost near-zero" against "review and verification capacity stays human-bound." The quadrant where build is free and verification is scarce tells you to invest in the harness before headcount.
LeadDev Engineering Leadership Report 2026 — Six hundred leaders report more hands-on work, longer hours, and lower team motivation; useful benchmark data when you make the case for realistic scope and manager support.
Deloitte 2026 Global Technology Leadership Study — 42% report minimal or zero AI ROI while 81% feel equipped to scale it; the confidence-versus-return gap is where program leaders earn their keep.
Revised rules of engineering leadership — Will Larson on individual-owned migrations and the development harness as the new leverage point.
"The greatest danger in times of turbulence is not the turbulence; it is to act with yesterday's logic."
— Peter Drucker, Managing in Turbulent Times (1980)
Don't miss what's next. Subscribe to Critical Path: