Most AI proposals you read this quarter quietly assume four things they will not write down. We wrote them down — as things we refuse to say. If a vendor's pitch deck depends on any of these, you have a fast disqualifier.
The four sentences we will not write
1. "Our system is general intelligence." It is not. Nobody's is. The current generation of large models are very good statistical pattern-matchers shaped by training data and human feedback. Calling that "general intelligence" is a marketing shortcut that quietly transfers risk from the vendor to you. Themesis has written carefully about the gap between today's "pseudo-persons" and what genuine autonomous systems would actually require — see How to Prepare for the AGI World: Tools, Pseudo-Persons, and Slaves (Class E). Our own framing on the science side is narrower and more honest: we are working on an attainable path toward General Natural Intelligence — natural, not artificial — and every claim we make about it is evidence-classed and falsifiable.
2. "This will heal you / lower your trauma / rewire your brain." Software does not do that. Trauma-informed design is a different thing — it means we build interfaces that don't ambush people, don't punish hesitation, and don't reward dissociation. That is a craft commitment, not a clinical claim. We will never tell you the product is therapeutic. If a vendor implies it is, walk.
3. "Guaranteed outcomes by week N." We do not sell certainty about your business. We sell a transparent build with an append-only ledger, named gates, a refusal-rate metric, and a written exit condition. The honest version of "what will week 12 look like" is: here are the artifacts that will exist, here is who owns them, and here is the falsifier that would prove the approach wrong (Class C). That is a contract you can actually audit. A guarantee is not.
4. "Anyone can do this — no technical ability needed." No. The work is learnable, but it is real work. An internal owner has to sit in the gate reviews, read the receipts, and exercise judgment. We will train your people; we will not pretend the training is unnecessary.
Why the refusals are the product
The reason this page exists at all is that the things we do claim only mean something if you can trust the things we won't. A receipt-first engagement model collapses the second a vendor inflates one claim. So we publish the no-list up front:
- No promise that the model "understands" you. It pattern-matches over your context.
- No promise that the system is autonomous. A human owner is in the loop, by design.
- No promise that we will out-perform an unspecified baseline. Baselines are part of the engagement.
- No promise that the ledger will be flattering. The ledger is the truth of the project; we do not curate it for slides.
Each of these is a disqualifier for a certain kind of buyer — the kind who wants a marketing artifact, not a working system. That is fine. We are not for them.
What we will say
We will say that we build with falsifiers attached to every claim, that our gates and override rates are visible to your internal owner, and that our exit condition is named on day one. We will say that the science underneath is published in the open under the UNI program, and that the agency work here at SolutionWright is the receipts-first translation of that science into practical client engagements.
We will also say, plainly, that this is a working hypothesis under continuous test. Don't take it on faith. Inspect the ledger, read the falsifier, and tell us where it breaks.
Read next
- The standing disclaimer page — the canonical short-form list, linked from every proposal we send.
- Transparency architecture overview — how the receipts, ledger, gates, and exit conditions fit together.
- The science behind the agency work — UNI program, evidence classes, published falsifiers.
- Book the workshop — the one-day session where we walk an internal owner through reading the receipts on a live project.