# Future of Engineering Work

#concept #ai

## Summary

How AI restructures engineering *labor and roles* (as opposed to the services *market*, which is [[ai-market-shift]]). From [[2026-07-06-sebastian-interview-ai-and-software-engineering]]: once "AI can write software" is settled, the whole game is *how you use it* — and that cascades into team size, tooling, the enterprise, open source, and identity.

## Current Understanding

**The premise:** "It's not a question *if* AI can write software anymore — it's just a question of *how* you use it." Cost of writing code trends to zero; leverage moves to direction and verification.

**Teams collapse from ~8 to 2–3.** No more scrum master + PM + requirements engineer + big dev team — instead one coordination/ownership role plus one or two people managing coding agents, sharing responsibilities. Coordination overhead can now exceed the work: on a small 2-person project Sebastian is **faster alone** than synchronizing who-does-what. (Cross-source: "team size is no longer a signal of seriousness" — the 4-person squad at ~$7M/yr in [[2026-06-15-rodenko-selling-development-expensively]]; solo-founder Rezi at $293k/mo in [[2026-06-15-making-money-with-ai-2026]].)

**Bring-your-own-harness vs company-managed — and the business opportunity.** [[eugene]]'s thesis: every developer should build a personal harness on top of Claude Code (his own: a Telegram-like UI, one agent per project, inter-agent messaging, per-agent memory, "done thinking" signals; tools like **Conductor** for git-worktree-per-chat). [[sebastian]]'s counter: compliance and liability make ad-hoc per-developer setups impossible at scale — "it has to be a company-managed resource." **→ open market: compliant, centrally-managed, company-standard harnesses for large regulated teams.** This is a [[niche-selection]]-shaped opportunity ("go where the indie tools can't").

**Enterprise reality is far more locked down than the indie world:** managed VMs, no personal laptops, zero self-installed tools; a ~1,200-engineer Roche SAP program; banks moving from banning AI to cautious adoption "because it's just so good."

**Two smaller theses:** open source will *grow* (near-free code is easy to give away; Eugene's cynical read: OSS is mostly marketing). Legacy/hobby niches persist (COBOL in banks — no training data; people who code for love "like an old-timer car") but not where time/quality/money matter.

**The philosophical turn — decouple identity from profession.** As many professions collapse into "prompt the AI," people who tie identity to their job title ("I *am* a doctor") will feel worthless; the advice is to separate *who you are* from *what you do*. "Fundamentally I'm Eugene — I'm not a programmer."

**Where the human value goes** is covered by the sibling pages: judgment ([[seniority-and-ai]]), ownership ([[product-ownership]]), and trust ([[relationships-as-moat]]).

## Evidence

- "It's not a question *if* AI can write software… it's just a question of *how* you use it." — [[2026-07-06-sebastian-interview-ai-and-software-engineering]]
- Team collapse to 2–3; coordination overhead > work; faster-alone — same source
- BYO-harness demo and Conductor; compliance counterpoint; the managed-harness market — same source
- Enterprise lockdown; Roche ~1,200 engineers; banks ban→adopt — same source
- OSS grows; COBOL/old-timer legacy niches — same source
- Identity decoupling — same source
- "Team size no longer a signal": squad model — [[2026-06-15-rodenko-selling-development-expensively]]; solo Rezi — [[2026-06-15-making-money-with-ai-2026]]
- **Counter-evidence (adversarial):** the "coding cost → 0" premise is qualified — AI's measured effect is modest (≈14–19%), can be *negative* for experts, and team-level delivery *stability* fell with AI adoption (DORA 2024) — [[ai-productivity-evidence]] / [[2026-07-18-ai-productivity-adversarial-evidence]]

## Related Pages

- [[ai-market-shift]] — the market-side twin of this labor-side page
- [[seniority-and-ai]] · [[product-ownership]] · [[relationships-as-moat]] — where human value migrates
- [[niche-selection]] — the enterprise-harness gap is a niche opportunity
- [[ai-productivity-evidence]] — the empirical qualifier on "coding cost → 0" and team collapse
- [[team-growth-ceiling]] — if teams collapse to 2–3, each person's growth rate becomes a larger share of the company's ceiling
- [[eugene]] · [[sebastian]] · [[virtido]]
- [[overview]]

## Contradictions / Uncertainty

- **BYO vs managed harness** is a live disagreement between the two speakers, not a settled point; the reconciliation ("personal for individuals, managed for enterprise") is the vault's synthesis. `Status: tentative`.
- **OSS motivation** is disputed within the source (Eugene: marketing; Sebastian: expects more).
- Team-collapse numbers (8→2–3) and "faster alone" are one founder's experience on small projects; Sebastian is explicit he doesn't know how this plays out on ~1,200-engineer programs.
- **The "coding cost → 0" premise is empirically qualified (2026-07-18).** [[ai-productivity-evidence]] finds AI's real effect is modest (≈14–19% average, negative for experts on familiar code), and — most relevant to *team* restructuring — DORA 2024 found AI adoption correlated with *lower delivery stability* (individual speed didn't convert to better shipping; DORA reversed the throughput half in 2025, stability persisted). This doesn't refute team collapse, but it undercuts the premise that the coding *itself* is now free and frictionless; the coordination and verification work this page centres on is exactly where the measured cost stays. Early-2025-scoped — see the source's time caveat.
- Identity-decoupling is philosophy, not evidence — included as a recorded view, not a claim.

## Next Questions

- What does a compliant enterprise harness actually require (audit, secrets management, standardization) — and is anyone shipping one?
- Does the 8→2–3 collapse hold on large programs, or only on small teams?
- If [[eugene]] is the vault owner, is the enterprise-harness market the business this vault should be scoping?