What does the tool actually decide?

It ranks seven solution shapes against what you describe: no AI, deterministic automation, search and RAG, an AI-assisted workflow, a copilot, a single agent, or multiple agents. You get the best fit, the two runners up, and the reason one leads the others.

Does my input leave my computer?

No. The model runs inside your browser, so the text you type stays on your device. There is no automatic tracking of what you write. If the tool fails to understand a sentence you can press a button to send that one sentence, and nothing is sent unless you press it.

Why is there a 36 MB download?

Because the model itself is downloaded and run locally rather than called over an API. It is cached after the first visit. That is the price of the text never leaving your device.

Is this a demo or a real decision aid?

It is the same reasoning used at the start of a real engagement, made self-serve. It narrows the shape and surfaces the governance questions. It does not replace the scoping conversation, it makes it start further along.

What if it does not understand me?

It says so and asks for more, rather than inventing a recommendation. A wrong confident answer would be worse than a prompt to say more.

AI Solution Shape: Decide What to Build Before You Build It

Most AI projects choose the shape first and justify it afterwards. Someone says “let us build an agent”, and six weeks later the team is debugging a multi-agent system for a process that three deterministic rules would have handled.

The shape is the decision that everything else inherits. Get it wrong and no amount of prompt engineering, model upgrades or budget recovers it.

The seven shapes

Every automation candidate lands in one of these:

No AI. Rules, or the status quo, done properly.
Deterministic automation. Predictable input, clear rules, no judgement.
Search and RAG. The answer already exists in your documents.
AI-assisted workflow. AI extracts or drafts, a deterministic step executes, a human owns the outcome.
Copilot. A person stays in the loop and asks as they work.
Single agent. One system, multiple steps, real autonomy.
Multiple agents. Several systems, genuinely separate responsibilities.

Most teams reach for 5, 6 or 7. Most problems belong in 2, 3 or 4.

Describe the situation, see the shape

The tool takes plain sentences. Your goal, the systems involved, what has to stay inside the EU, how many engineers you have, whether accuracy matters more than speed. It captures each signal as you type and ranks the shapes against them.

It also tells you when it disagrees with you. Ask for a copilot and it will still rank a copilot, but if what you described is really an extraction step with a review gate, it will say so and show you why. A tool that only confirms your assumption is not worth the download.

Open the AI Solution Shape tool

What you get

A ranked recommendation with the reasoning, plus the detail layer that usually takes a workshop to produce: the integration spine from source system to system of record, who owns a record once an agent creates it, how it is labelled as agent-created, and how it gets rolled back.

That last part is the part most pilots skip and most production deployments fail on. Governance is not a phase after the build. It is part of the shape.

You can download the result as a PDF, or share a link to it, and bring either one into a scoping call so the conversation starts from a diagram rather than a blank page.

It runs on your machine

The model is a 26 million parameter function-calling model, downloaded once and run inside your browser. Your text is never sent anywhere. That is a deliberate constraint: a tool that asks you to describe your internal processes should not be quietly collecting them.

It also means the tool is honest about its limits. It is small, it is fast, and when a sentence does not land it asks for more rather than guessing.