A thinking layer for the AI builder era

Context collapse

The industry is full of prototypes cosplaying as products. Polished screens, smooth interactions, and no memory of who they are for or what happens when something goes wrong. That is not a craft problem, it is a context problem: an AI builder can hold your words, but a one-paragraph prompt cannot hold your research, your edge cases, or the decisions you already made. The model fills the gaps with its best guess, and the guess is confident.

Tessera is the bet that the fix belongs upstream of the prompt. Instead of flattening product thinking into a document, it keeps the thinking as a connected model: personas, goals, problems, research, flows, tasks, screens, decisions, each linked to the inputs it came from. The agent receives the graph, not a summary of it.

The conversation is the interface

The kickoff is a structured conversation, not a chat. Tessera reads everything you bring, then works through the model one question at a time. Each card shows its reasoning in full: what it read, what it weighed, and the option it leans toward, labeled as a lean and framed to be pushed back on. You can pick an alternative, correct an earlier answer, or say I don’t know and keep moving.

The right rail keeps score honestly. The model outline builds as you talk, and every row is tagged with how it got there: discussed, from you, or inferred. Progress reads as coverage of the model, not length of the transcript.

One model, many hands

The workspace is where the co-authored model becomes yours to shape. Cards for personas, problems, tasks, and flows sit in one grid with three ways to read it: compact, expanded, and a map. Filters surface what needs attention first, open issues and inferred fields, and the thinking partner stays docked on the right for questions and changes that ripple through the model.

Where every field came from

Most AI tools are black boxes; Tessera is tagged. Every field in the model carries its source: you said it directly, it came from a document you uploaded, it emerged from the conversation, or the AI inferred it. From you reads green, inferred reads sky, so you can scan a card and see exactly what you authored and where to look closer before you build on it.

The wrong thing first

The graph view was the first thing I built and the thing I was proudest of. My first user opened it, went quiet in the unproductive way, and I knew: it was trying to be an editing surface, and they could not tell what editing meant there. I pulled editing out and made it read-only. It survives today as the Map toggle in the workspace: orientation, not authoring.

From model to build package

Generation and review are two passes, not one. A first pass drafts, a second pass critiques what the first one made, and anything flagged routes to you. Nothing writes itself into the model unreviewed, which is what makes the model trustworthy enough to export.

The export is a build package, not a copy-paste document. It carries the model summary and a readiness score, with starter prompts tuned to where the code gets written: Claude Code, Cursor, Codex, or a copy target like Bolt, v0, or Lovable. The thinking is done before the building starts, which is the point.

Outcomes

Reflection

I spent weeks building the graph as the primary editing surface before putting it in front of anyone. One quiet user test undid it in twenty minutes, and it was the right undoing. The lesson I keep relearning: the thing I am proudest of is usually the thing I need to pressure-test soonest. The name keeps me honest too. A tessera is one tile of a mosaic; every product is a mosaic, and the work is how the tiles hold together.