The starting point was not AI. It was Bitcoin. Don't trust, verify — and nobody can touch the ledger. I ran my own nodes, held my own keys, lost sats in broken hardware. Then I took the same rule off money and put it on everything else the platform was holding for me.
Same move, one layer up each time. Not your keys, not your knowledge.
It started with losing projects. Everybody is good enough now. So I stopped arguing with that and built the thing that keeps me relevant.
Source: Recorded meetup conversation, Portugal, May 2026 · own posts + build history, 2025–2026 · recorded conversations, AI Summit Lisbon, June 2026 [raw-check before publish]AI's big fault is that it always compresses. It summarizes, it compresses. I record lossless — like music. Not MP3, not MP1. Every transcript, everything that happens, stays raw on my own SSD. When better reasoning arrives, I run the recording again and it finds new connections. Nothing was thrown away.
The reason it has no hallucinations is not a feature I trust. It's a rule I enforce. Nothing counts as verified unless it carries its source and an exact quote from that source, and the quote still matches the source word-for-word. No source, no marks, no match — it stays labelled as unverified. Knowledge needs proof of fact: prove the data is real from the ground up, not some hallucination.
And the last check is not the machine. Only the expert can audit the output. Ask an architect to build you a nice house from a prompt and everything is wrong. I can't be in your shoes — the expert has to be.
Source: Recorded meetup conversation, Portugal, May 2026 · the system's own verification rules, git-tracked · recorded conversation, AI Summit Lisbon, June 2026 [raw-check before publish]Opens — my readWhat does "verify" mean for the thing you're building? Where's the ledger nobody can touch? Pulls the crypto-native and the security people straight into architecture.
Say: recorded meetup conversation, Portugal, May 2026Opens — my readWhat are you losing every time the model summarizes for you? Good bridge to the RAG and eval crowd — what gets dropped before retrieval even starts.
Say: recorded meetup conversation, Portugal, May 2026Opens — my readWho audits your agent's output, and do they have the domain in their hands? Direct line into tonight's eval-framework talk — vibe testing versus expert audit.
Say: recorded conversation, AI Summit Lisbon, June 2026 [raw-check before publish]Opens — my readFounders answer this differently than engineers. It moves the table from tools to moats without me having to say the word "moat."
Say: first half recorded at AI Summit Lisbon, June 2026 [raw-check before publish] · the question is framing, not a quoteOpens — my readWhose frame are you building on top of, and who moves the terms when it matters? Pulls the platform and distribution people — a bridge to "distribution is a system."
Say: recorded conversation, AI Summit Lisbon, June 2026 [raw-check before publish]Opens — my readWhat problem are you actually solving before the build? Minimal usable, not minimal viable — least effort in, and you already recognize if it's useful. Pulls product people off the feature list.
Say: two recorded conversations, AI Summit Lisbon, June 2026 [raw-check before publish]Opens — my readInvites the builders to describe their own loop in plain language instead of a stack diagram. The music frame does the disarming.
Say: recorded conversation, AI Summit Lisbon, June 2026 [raw-check before publish]