8 November 2026
I didn't build a dashboard
When you build enterprise software, there's a thing everyone expects you to build next: the dashboard. A portal. A screen you log into, with a left-hand nav and a row of tiles and a search box. It's so automatic that "where's the UI?" is usually the first question people ask about the compliance system I've been building.
The honest answer is: there isn't one. And that's deliberate.
Where a compliance answer actually needs to appear
Think about when you actually need to know whether you're compliant with something. It's almost never while sitting in a dedicated compliance portal. It's in the middle of other work — reviewing an architecture, scoping a release, answering a regulator, writing a policy. The question arrives in the flow of something else.
A separate dashboard asks you to stop, context-switch, find the right tab, remember your login, and translate your real question into whatever that tool's search box understands. Most people don't. So the expensive portal sits unopened and the question gets answered the old way — someone's best guess, in a hurry.
So I made a different bet. Instead of building a place you have to go, I put the system where the work already happens: inside the AI assistant your team already uses. It installs as a plugin into Claude Desktop, Claude Code or VS Code — a set of skills (ask a question, ingest a regulation, author a policy, clean up duplicates in the graph) backed by a server that reaches the governed compliance graph behind the scenes. You ask, in plain language, in the tool you already had open. The answer comes back there.
No UI is a forcing function for honesty
Here's the part I didn't expect, and now wouldn't give up. Having no interface of its own forces the system to be honest.
A dashboard gives an AI somewhere to hide. It can present a confident-looking screen — charts, a green tick, a number — and you'll tend to trust the furniture. With no UI, there's nowhere to dress up an answer. The assistant has to say, in words, exactly what the graph returned and where each claim came from. The queries underneath are deterministic; every factual claim has to trace to a specific source in the data; the answer is checked against the graph before it's given, and if the graph can't support it, the system says so rather than quietly filling the gap from memory.
That's the whole philosophy of the thing in one design decision: the AI is the interface to a verifiable source, not the source of truth. It writes the query and phrases the answer. The graph — live, cited, deterministic — is what's actually authoritative. Stripping out the UI makes that impossible to fudge.
The two places I didn't manage to avoid a screen
I'll be honest about the edges, because the brand here is honesty. It isn't literally pure chat.
There's a thin command-line tool for a handful of operator chores — health checks, pulling an export, setting up credentials. And there's exactly one old-fashioned web page the system serves: the signing page. When you do something irreversible — like merging two obligations in the graph — committing it takes a passkey signature, the same fingerprint-or-face gesture your phone uses, and a browser is the only place that ceremony can happen. I've written before about why some decisions should cost a fingerprint; that one page is where the fingerprint gets pressed. Everything else lives in the assistant.
Could there be a richer interface one day? Maybe — there's a design on the drawing board. But the default I'm building from is agent-first, not dashboard-first, and that ordering is the point.
What this means on Monday
If you're evaluating or building an AI tool, ask a question before you ask about features: where does the value actually need to show up? For a surprising amount of software, the honest answer isn't "a new dashboard my team has to adopt." It's "inside the tool they already live in, at the moment the question arises."
Meeting people where they already are beats building somewhere impressive and waiting for them to come to you. That's true of software. In my experience it's just as true of leadership.
I'm Tete Mensa-Annan. I help engineering and compliance leaders adopt AI in a way that's honest about what it does and doesn't know. If your team is drowning in portals nobody opens — or weighing whether to build another one — I'm happy to talk it through. Not a pitch, a genuine conversation.