What six months of building and validation changed about our thesis, product pivots, category positioning, and roadmap reality.
Earlier this year I published a page that said, in effect: conflict is not failure, it is signal, and organisations have no infrastructure to make it legible. That page had an owl on it, a five-role visibility model, three resolution pathways, and a seven-phase roadmap that stretched to day 180.
Most of the thesis survived. Almost none of the mechanics did — at least not in the shape I drew them.
Here is what actually changed, and why.
The old page called this infrastructure. Infrastructure for making conflict legible. Accurate, and completely inert — the word invites you to imagine plumbing.
Entellispace is emotionally intelligent workspaces, and the claim underneath it is simple: emotionally intelligent organisations handle conflict better.
That is not a softer way of saying the same thing. The original framing put the intelligence in the system: it extracts, it classifies, it routes. The current framing puts the intelligence in the workspace — the shared thing a team actually inhabits. The system's job is to give a workspace the capacities a good manager has and a Slack channel doesn't: noticing when someone said yes and meant no, knowing which disagreement needs air and which needs a decision, remembering that this argument is the fourth time this quarter.
It also fixes a gap in the old page. That page was good at detecting conflict and vague about what makes a team any better at holding it. The answer isn't a routing table. It's the emotional capacity of the organisation itself — and that is a thing you can build for.
Read the original page again with that lens and the five-role visibility matrix looks like what it was — an org chart pretending to be a product. Emotional intelligence isn't a permissions model.
Somewhere in here Intellispace also became Entellispace.
The original page opens with a PM pasting a transcript. Every screen, every role definition, every pathway assumed the PM was the centre of gravity.
The PM is not the buyer.
The PM is the person who feels the pain most often and has the least budget to fix it. The buyer is the time-pressed leader — the one who finds out three weeks late that a decision everyone thought was made was never actually made. That person doesn't want a decision queue. They want to stop being surprised.
This changed the product more than any technical decision did.
The original classification was clean: decided, unresolved, structural. Three buckets, three pathways.
Clean was wrong. Real transcripts don't sort into three buckets — a single meeting contains a decision that's actually a deferral, a tension that's really a resourcing constraint, and a structural signal disguised as a joke about the roadmap.
The classifier now runs a twelve-dimension decision_type model, and decisions carry project_id linkage so a trace can survive across meetings instead of dying at the end of one transcript. The three-bucket page was a description of the idea. Twelve dimensions is what the idea costs.
The page says PASTE → transcript. That's still there. But the thing that actually gets used is voice capture straight in the browser — MediaRecorder into a Gemini transcription endpoint. No copy-paste ritual, no waiting for the meeting tool to email you a file.
And the hardest, most valuable problem turned out to be one the original page never mentioned: messy multi-thread transcripts. Four people talking over each other about three topics. Every prospect asks about it. It is now the differentiator, not an edge case.
Three pathways — decision trace, threshold negotiation, HR escalation — were the right instinct at the wrong altitude. They described what the system does rather than what coherence requires.
What replaced them is the Coherence Protocol: Sense, Surface, Steer. Sense what's actually in the room. Surface it to the right audience, in the right register, in the right sequence. Steer toward resolution. The three pathways still exist inside it — they're implementations, not the frame.
That framework also travelled outside the product, into a workshop suite under the Reframe brand. Turns out you can teach the protocol to a team before you sell them software that runs it.
Per-person conflict routing was in the original design. It requires knowing who is on what team — pod membership.
I started building it. Then I stopped mid-flight.
Not because it's wrong, but because a pod is structurally just a design-partner team, and I hadn't validated that anyone wanted routing granular enough to need it. The bottleneck was never building. It was validation. That's the single most expensive lesson of the last six months, and the reason a half-finished feature is sitting where it is instead of being finished on principle.
After shipping the Product Hunt debut, the sudden traffic spike felt like a success. The launch brought [TK: upvote count] upvotes and [TK: signup count] signups on [TK: launch date]. But I quickly had to learn to treat the debut as a distribution event, not a validation event. High traffic is not evidence that anyone actually has the problem you are solving; it is just a measure of temporary internet attention. It is easy to mistake pageviews for validation.
To find real evidence, I scheduled and ran eight customer interviews with [TK: profile of interviewees]. They took weeks of coordination, but they yielded the weakest signals of all. In retrospect, I fell straight into the trap of collecting opinions rather than facts. Rob Fitzpatrick's Mom Test framework defines the rule clearly: talk about their actual life, not your idea; focus on past behavior over future intention; and remember that anything short of a active commitment of resources is just a compliment.
I had entered those interviews with questions that were classic compliment-generators, asking things like [TK: insert compliment-generator question] and [TK: insert another compliment-generator question]. Because people are naturally generous to solo founders, they gave me polite, theoretical encouragement. But none of it represented a real commitment of time, money, or reputation. They said they would use it, but they had no skin in the game. Real validation requires a transaction.
What actually validated the product was when a design partner ran a real, raw transcript from their own internal team meeting through the engine. That single session outweighed all eight interviews combined. The partner wasn't reacting to my pitch or looking at a clean hypothetical demo; they brought their own active operational mess. They wanted to see if the engine could make sense of their actual conflict.
The consequence of that single session was immediate. It is the direct reason why the pods feature stayed unfinished—I saw that they didn't need complex routing permissions yet. It is also why solving messy, multi-thread transcript parsing became my top priority. The decision to pivot didn't come from a predefined roadmap, but from watching a user hand over their real data unprompted.
The owl was good. Source Serif, Jost, IBM Plex Mono, cream and ink and coral and lavender. It had warmth. Every screen wore it.
Then feedback landed: this needs visual polish before it goes to investors. So I hardened the owl palette for enterprise — serif for marketing only, owl reduced to a static mark.
And then I threw that out too. The brand now is deep blue and orange, glassmorphism, Bauhaus geometry, a dark navy signal-map panel. The line under it: See the signals. Build the space for everyone to thrive.
Two brand rebuilds in one quarter is not efficient. But the enterprise buyer reads visual language as a proxy for whether you'll still exist in eighteen months, and that's a fair thing for them to read.
The original page had seven phases with day counts. Phase 4 was signal sources at days 60–90, Phase 6 was agentic architecture at 120–180.
The real list looks nothing like that. It's dual-processing assessment of how decisions get made, compliance checks against HR and legal constraints, which constraints can be lifted for innovation work, communication matrices, adjustable governance cadence, an ICP document, a stakeholder buy-in narrative, and the evals-and-harness work that any of this depends on.
Notice how much of that is not features. Phased roadmaps published on a landing page are a genre of fiction. Useful fiction — they tell people you have a direction — but fiction.
Contribution tracing. State the expected outcome of a decision at the moment you make it, then check in later on whether it was observed. Scoped only to decisions flagged high-stakes at creation — running it on everything would just be surveillance with extra steps.
A paid diagnostic. Short questionnaire, single root-cause report, one price. Not scoped, not built. But it's the shortest path between "I think my org has a coherence problem" and someone paying to find out.
The two numbers.
Decision velocity — how fast a team gets from conversation to committed decision. Decision clarity — what percentage of decisions were understood and acted on correctly. Thirty days to the first report.
Those were on the original page and they're still the promise. Everything else — the name, the buyer, the classifier, the brand, the roadmap, the pathways — turned out to be a hypothesis about how to get there.
The product is live now. There are beta testers, design partners, and one validation session on a real transcript that told me more than eight customer interviews had.
Conflict is still not failure. It's still signal. What changed is where I think the intelligence has to live — not in a system that watches a team, but in the workspace the team is already inside.