800 million ChatGPT users by 2023. Zero statistical productivity growth in business across four Western economic zones. Where did all the saved time go? TikTok, Instagram, entertainment.
This isn’t a bug. It’s a feature of how we manage today.
I put together an “AI Management Manifesto” - modeled after the classic Agile Manifesto. 4 values, 8 principles. Every point is grounded in real mistakes - mine and others’.
Four Values
We are constantly discovering better ways to build products with AI - by doing it ourselves and helping others do the same. Through this work, we’ve come to value:
Psychological capacity and tacit knowledge of the team over trendy AI tools and metrics.
Careful planning and rigorous verification over raw code generation speed and blind vibe-coding.
Async management of multiple AI agents over the familiar, focused, synchronous development flow.
Voluntary bottom-up experiments and sandboxes for enthusiasts over top-down mandated AI adoption under threat of layoffs.
While there is value in the items on the right, we value the items on the left more.
Eight Principles
1. Planning pays back tenfold
One minute spent writing a detailed spec for an agent saves 10 minutes untangling cascading bugs. An agent can run autonomously for 1.5 hours - but only with a perfect brief. No spec? You get vibe-coding with a predictable ending.
2. AI is a catalyst, not a rescue
If your team already had overload, distrust, and blurry roles before AI - AI will amplify that chaos at incredible speed. Had overload? AI floods the pipeline with incoming tasks, turning the manager into a meat grinder. Had distrust? AI call monitoring kills psychological safety. Garbage in, garbage out. Fix the sociology first, then add the technology.
3. Humans are the bottleneck
When agents speed up code writing, remember the Theory of Constraints: the next stage gets proportionally more work. AI triples code generation speed - the reviewer gets 3x more Pull Requests. If you’re not measuring end-to-end metrics (Lead Time: idea to production), the reviewer simply burns out. Per-stage metrics are dead.
4. Protect the carriers of tacit knowledge
AI agents can’t extract what isn’t written down. Business context, the reasoning behind architectural decisions made three years ago, political agreements - all of that lives in specific people’s heads. You can’t fire them. Their job is to systematically convert that context into formalized specs for AI agents.
5. Ownership only comes from choice
Pressure from leadership and imposed KPIs around AI usage lead to covert sabotage (31% of employees) and pretty PowerPoint reports. Build sandboxes, find enthusiasts, give them tools and time. Don’t pile on new tasks. Rely on organic growth through AI champions.
6. Writing code is dying - long live engineering
The classic SDLC is dying. Stages are collapsing into a single agent window. An engineer’s work is migrating to the edges: preparing requirements (input) and rigorous validation of results (output). New efficiency formula: Planning + Verification > Coding.
7. Teams only learn through their own practice
Having an external consultant craft perfect prompts doesn’t work. The tool becomes a “magic artifact” nobody actually uses. People need to make their own mistakes - solo experience is useless, you need a shared knowledge base and prompt exchange inside the team.
8. We accept the Productivity Paradox
Deploying AI doesn’t automatically improve overall business productivity. If a manager doesn’t restructure processes, the time saved by agents gets spent - guilt-free - on TikTok and Instagram. Process drives the derivative. Tools give a fixed delta.
What to do with this
Four steps for tomorrow:
- Legalize sandboxes. Find enthusiasts, give them tools and time - not new KPIs.
- Switch to end-to-end metrics. Measure Lead Time from idea to production, not code writing speed.
- Add transparency. Implement automatic export of AI work sessions for review.
- Watch the battery level. Don’t introduce AI when the team is deep in overload. Engage through voluntary choice.