Org design in the AI era: span of control is doubling.

Manager workload is roughly proportional to: (IC questions per week) + (review work) + (1:1 hours) + (career conversations). AI absorbs the first category dramatically — policy questions, how-to questions, second-opinion questions. It compresses the second (less to review when AI drafts the first version). The third and fourth don't shrink, and arguably should grow. Net effect: the same manager can support significantly more ICs if the work is redesigned. But only if.

1. 'Flatten' meaning 'remove one layer at the top'

Most 'flattening' announcements eliminate the VP layer and call it done. The real opportunity is in the M2/M3 layers, but it's politically harder because that's where most of your senior people sit.

2. Wider spans without rebalancing 1:1 quality

Going from 7 reports to 11 without changing how 1:1s work means each report gets less time. The math demands tighter 1:1 cadence, shared agenda templates, and a willingness to skip a 1:1 when there's nothing to discuss. Most managers will refuse to skip without explicit permission.

3. Keeping the same comp bands for IC and M1

If your most senior IC makes less than your weakest manager, every senior IC who wants more money becomes a candidate manager. That was a bad bug at span 7. At span 11 it's existential. Fix the IC ladder first.

Parkinson's Law — work expands to fill the time available — applies brutally to management work. A manager with 6 reports invents enough oversight, coordination, and 'alignment' to fill 40 hours. The same manager with 11 reports doesn't suddenly become twice as efficient; they just stop doing the lowest-leverage half of their old job. Whether that's a feature or a bug depends entirely on which half they drop. Without explicit prioritization, they will drop the wrong things — coaching, career conversations, written feedback — because those are the ones with no immediate stakeholder banging on the door.

Layer on Peter's Principle: people get promoted to their level of incompetence. When AI compresses IC work and you respond by promoting your best ICs into management to absorb the headcount, you double-tax the org — you lose their IC leverage AND inherit a population of managers who never wanted the job and were never trained for it. The 'span of control doubles' insight only pays back if you also build the IC ladder that lets your best builders stay builders.

An AI-native data startup founded in 2023 hit 80 engineers with 6 EMs (span ~13) and no director layer. They didn't add layers; they invested in three things: a written manager operating model with default 1:1 cadences by report seniority, a principal IC ladder with comp parity to Director, and async-default communication so the 11-report manager wasn't drowning in synchronous time. At their 18-month engagement review, EM eNPS was +52 and senior IC retention was 96%. The same headcount at a traditional company would have spawned 3 director hires.

• Review span every 12 months — it's a moving number, not a cultural norm.
• Build a principal IC ladder with comp bands matching Director and VP. Publish it.
• Tier 1:1 cadence: weekly for new hires and underperformers, 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 and different skills.
• Audit your last 8 senior promotions. If all went to EM, your IC ladder is performative — fix it before announcing wider spans.