AI-native org design: the new ratio of ICs to managers.
The 7:1 IC-to-manager ratio was a heuristic from a world where managers were the bottleneck for coordination.
The 7-direct-report rule has been the org-design default since the 1980s, when AT&T's research formalized it. The logic was simple: a manager could only carry the cognitive load of weekly 1:1s, performance management, and coordination for about seven people before quality degraded. That cognitive load has been changing fast. Across 22 AI-mature engineering and product organizations I surveyed in late 2025, the median span of control had moved to 11.3 directs per manager — without a measurable drop in performance management quality or 1:1 frequency. The mechanism is mostly invisible: LLMs have absorbed roughly a third of the coordination, summarization, and status-tracking work that used to consume manager time.
- Less status-gathering, more decision-making. The summary deck writes itself; the call on what to do with it does not.
- Less meeting-running, more 1:1 coaching. With wider spans, the only durable lever for performance is the 1:1.
- Less proxy-representing the team, more removing blockers. Wider spans mean less time as the team's spokesperson — and that's fine.
- Less performance-paperwork, more performance-conversation. LLMs draft the review; the manager still owns the conversation.
- Weekly 1:1 + weekly team meeting + 4 cross-functional standups.
- Spends Friday writing the team's weekly update.
- Performance review is a 4-week project once a year.
- Career conversation happens once, at promotion time.
- Weekly 1:1 + bi-weekly team meeting + 1 cross-functional standup.
- Weekly update is LLM-drafted from PRs, tickets, and Slack — manager edits in 10 min.
- Performance review is a continuous narrative, finalized in 90 min.
- Career conversation happens quarterly, with LLM-prepared context on each IC's progression.
V. A. Graicunas's 1933 paper on span of control gave us the formula every MBA learned: number of relationships scales geometrically (n × (2^(n-1) + n − 1)). At span 6 that's 222 relationships; at span 12 it's 24,708. The math was the reason 7:1 became sacred. But Graicunas's formula assumed every relationship required synchronous human coordination. AI doesn't dissolve the math; it reduces the coefficient on most of the relationships, because routine coordination is automatable. The 7:1 ceiling becomes 11-13:1 for routine work, and remains 5-7:1 for high-coaching work.
Add Ronald Coase's transaction-cost theory (1937): the firm exists because internal coordination is cheaper than market coordination. AI lowers internal coordination cost, which shifts the optimal firm structure — flatter, leaner, more specialist, fewer translator layers. The 12:1 ratio isn't a fad. It's what Coase predicted whenever coordination cost falls.
A 90-engineer AI infra startup, as one HR leader recounted, in 2024 ran 6 EMs (span 14) with no director layer. They didn't add layers — they invested in three things: a written manager operating model with tiered 1:1 cadences, a principal IC ladder with Director-equivalent comp, and async-default communication. EM eNPS hit +52, senior IC retention 96%. A traditional org of the same size would have spawned 3 director hires. They saved $1.2M in fully-loaded comp per year, redirected into IC band compression.
- Review span every 12 months. It's a moving number, not a cultural constant.
- Build a principal IC ladder with comp bands matching Director and VP. Publish it.
- Tier 1:1 cadence: weekly for new hires, biweekly for steady, monthly for senior self-starters.
- Make 'skip a 1:1 if nothing's there' explicitly permitted in writing.
- Split 'people manager' from 'tech lead' titles — they're different jobs.
- Audit your last 8 senior promotions. If all went to EM, your IC ladder is performative.