We rehearsed our own positioning before we rewrote the site.
creativehuman.ai is rebuilt. The argument now runs in the order clients meet us in: advisory, upskilling and custom builds first, Glasshouse and Brew Chai underneath as the reason any of it holds. We built it the way we build everything else, and we put our own go-to-market question through Glasshouse eight days before it went out the door.
Creative Human AI · 30 August 2026
The old site led with the technology and left the reader to work out what we would do for them. That is the wrong order for a lab that sells judgement. The rebuild puts the three things you can engage us for at the top of the page, and the two platforms that make them work underneath.
What changed
- Services first. Advisory, Upskilling and Custom each have their own page, with what a session or a track or a build actually looks like, and what you walk out holding.
- Glasshouse has its own page. One run, end to end: you tell it the decision, it drafts the scenario, the cast assembles, the room argues over timed rounds, and the reading comes back with its receipts.
- Three named tracks. Leadership, Executive and Engineering, with the format written out. No more "we also do training".
- The record is in one place. Field notes carry releases, platform openings and client engagements as they happen, this one included.
- Contact is the email. The address is set, the subject is proposed from your words and confirmed by you, and the rest is what you write.
Built the way we build
The site runs on Brew Chai, the same orchestration layer we use on Glasshouse and on client work. One command, one PR, tests included. Every page is generated from source rather than hand-edited in place, which is how the instrument pages, the archive pages and the room pages hold one type system, one colour language and exactly one easing curve between them. The Brew Chai repository opens shortly.
What the run said
On 22 August we put our own go-to-market question to Glasshouse in the form we would put a client's: "Which buyer group should we go after first — early-stage founders, mid-size and large organisations, or investors — and what do they object to and what changes their mind?" Ninety-four personas across X, LinkedIn and Reddit. Nineteen of thirty-five scheduled rounds, stopped by its own budget once it had converged. 2,869 posts and actions. Run cca0e2ff.
It came back recommending mid-size and large organisations first, ahead of early-stage founders and investors, at HIGH run confidence. Its own title for the reading: "Target mid-size organisations first — they have budget and demand auditability."
That confidence is worth reading properly. Glasshouse reports a run's confidence as the weakest of the inputs that count towards it, never a blend of them. Grounding came back at 92 of 100, fidelity at 92, sample size on 2,869 contributions. All three HIGH. Dominant-frame alignment came back LOW, at 20, and the product prints it next to the tier while deliberately keeping it out of the tier. The inconvenient number is on the page, not folded into the good ones.
The same care is owed to the headline count. The framing that won carried 572 posts across 93 authors and all 52 cohorts in the room, which measures how much the room discussed it, not how many agreed. Inside the winning cluster the stance counts run 92 positive against 184 negative, and the reading flags that split itself.
Then the decision table docked its own evidence. The row it recommended cited four posts for its revenue range. One of the four resolved to no post in this run, so the table dropped that citation, let the range stand on the surviving three, and wrote the drop down rather than ship a pointer to nothing. That line is not the model narrating itself. It is a log format string, and its two numbers are counted off the citation list before and after the drop. A date would only prove the code could have run. Numbers filled in by counting prove that this sentence came out of it.
The guard is narrower than it sounds, and the narrow version is the one worth publishing. It walks the decision table's rows and nothing else. Six of this run's twenty-six citations would fail the product's own resolution test, and every one of the six sits in a section the guard never visits. The run is one day older than the check that closed that gap. We quote none of the six, and the line we quote below was matched against the stored post character for character, speaker included, before it went in.
The room argued shared infrastructure against dedicated tenancy between rounds fourteen and eighteen. It never reached the verdict, and we are not going to write it up as though it had.
The line the reading chose to lead with came out of round one.
Mid-size organisations are the only buyer group worth sequencing first. They have budget to spend but lack the procurement muscle to catch hidden costs.
We took the sequencing, and we took the objections attached to it. The reading is specific about what changes a mid-size buyer's mind: documented miss rates, EU-only inference guarantees, auditable methodology. Those are things to show rather than say, and showing them is what the Glasshouse page is now for.
The full account of how we work is on the lab page. If you have a decision worth rehearsing, bring us a scenario.