Let there be impact. Fast.
What we believe. How we behave.
Our view on AI: Do we think too much? And feel too little? →
Values are what we'd defend when it costs us. Principles are how we act on a Tuesday afternoon under deadline. Both are documented from practice, not written for a wall.
“To understand anything — zoom out.
To change everything — zoom in.”
People in control
AI assists, never dictates. Every automated decision is overridable by a human, by design. Agents do the work; humans orchestrate.
Reality first
Reality is the only dashboard that never lies. We capture it, make it queryable, and simulate against it — instead of adding another proxy system on top of forty years of system fat.
Explainable everything
Retrieval over generation. Physics over statistics where it matters. If a system can't show its reasoning — or say "I don't know" — it doesn't ship.
Zero lock-in
Every component replaceable: cloud, model, language, vendor — including us. We architect so you can fire us and keep everything working.
Private by default
Edge-native, local-first, sync-later. Data sovereignty is architecture, not a policy PDF. Personal data never leaves the person.
Simulate before you touch reality
Every rollout, parameter, treatment and build is rehearsed virtually first. Failure belongs in simulation, where it costs a rerun instead of a subscriber, a shutdown, or a life.
Ship — or it didn't happen.
First users beat feature lists. Something real by end of day beats something perfect next month.
Never wait.
Waiting is a banned word. Best-case assumptions, start anyway, escalate within minutes — not days.
Bring options and a default.
No open questions. Three plans, a recommendation, and "we go with D if you don't object."
Challenge the plan — anyone's.
Including ours, including the AI's. Unchallenged plans are how teams build on hallucinations for weeks.
Robust. Useful. Attractive.
Modular enough to replace anything, useful enough to matter, clear enough that the next human — or AI — can work on it. Useful wins.
Deterministic where it counts.
High-stakes systems answer the same way a thousand times. 8% of the stack is AI; 92% is the scaffolding that makes it trustworthy.
Never figure it out alone.
Trees in a forest are never alone. Thirty people silently stuck on the same problem is months of life burned.
Test it before you show it.
"Installed" is not "working." Demos over slides. Claims carry evidence or they carry the label "assumed."
Extreme Collaboration.
Repeatable, understandable, explainable, every step toward autonomous operations. Forming and performing, without the norming.
These aren't aspirations — they're quotes with timestamps. Each principle traces to something said out loud, on the record, while running a 30-person program screened from ~300 applicants over two months. The working culture is documented in the Amity Codex.
Extreme challenges need Extreme Collaboration.
Doctrine, not decoration.
The values run through everything we ship — from telecom twins to Life Atlas. The principles are enforced daily, in standups, reviews and deploys. When they conflict with convenience, the principles win.