How an analyst and Claude share a .deepcell

Halfway through a build, the question shows up: am I supervising the agent, or driving? The model is taking shape. The dashboard is filling in. Some of the cells you wrote; some Claude wrote. By Friday you'll send this to a partner with your name on it.

Where exactly is the line?

The handoff most demos hide#

Most agent demos cut away from the seam. The prompt goes in, the deliverable comes out, the analyst nods. It looks like collaboration. But "Claude built this" and "I built this" are not the same artifact, and the difference is what determines whether you can sign your name to it. If the line between the two isn't legible, you didn't share the work. You ceded it.

The fix isn't more guardrails on the agent. It's a shared surface where every move is visible — to you, in the moment, and to anyone reviewing later.

The file is the surface#

A .deepcell is that surface. The analyst opens it in the browser playground, or works with it through a set of typed commands. Claude opens the same file through a secure connection. Both of you read and write the one document — the same line items, the same formulas, the same linked calculations — not two separate copies. No mirror, no proxy, no "the agent's working copy." One file.

Every interaction is captured twice. During the session, you watch each move the agent makes as it happens. After the session, every edit is a saved version in the file's history. You don't have to ask what did the agent do? — you can ask the file to show exactly what changed, a line-by-line before/after comparison.

How the wiring actually works#

Connect. Claude.ai, Claude Code, or any assistant you use connects to your DeepCell workspace through a single point of entry that opens the whole toolset. On your own machine, the connection is local and needs no sign-in. From anywhere else, it uses the same kind of secure sign-in your banking app uses, plus a personal access key you create and manage in your account settings.

Delegate. The orchestrator is Claude Sonnet 4.5. It hands bounded jobs to three specialized helpers, each driven by a complete per-task brief:

• A general research-and-modeling helper — executes research, modeling, and report briefs, with web search, file search, the format guide, and the tools to build the model.
• A filings extractor — pulls filings from SEC EDGAR and categorizes line items.
• An Excel importer — converts a spreadsheet into a .deepcell (see bringing your Excel model).

None of them improvises beyond its brief and tool surface. The Excel importer doesn't rewrite your formulas; a modeling brief doesn't go off to scrape a 10-K it wasn't asked for.

Edit. A short exchange, the kind you'd actually have:

You: Add a sensitivity table for WACC ± 100bps and terminal growth ± 50bps.

Claude: Done — I added the sensitivity grid, with each cell wired to the inputs it depends on, so it recalculates on its own. That's one new line item, one new formula, and the twelve values that follow from them.

You didn't have to inspect the formula bar of twelve cells, and you didn't have to take Claude's word for it. Ask the file what changed and it hands back the same short summary — one new line item, one new formula, the values that fall out of them. If anything looks wrong, you roll back to the version before and ask again.

What the agent cannot do#

This is the part that matters most, and it's the part no AI-inside-Excel story can offer.

The agent is walled off from a specific set of actions: it can't export a spreadsheet, can't send anything outside your workspace, can't sign in as you, and can't publish a change to a shared copy. The principle is consistent — anything that reaches outside the file, or commits you to something, stays with the analyst.

The agent can read your file. It can query values. It can edit values and definitions. It can trace the reasoning behind a number — which assumptions support it, which evidence contradicts it. What it cannot do is publish a change to a shared copy, hand an exported spreadsheet to a stakeholder behind your back, or log in as you.

Blocking the Excel export is deliberate. Exporting a file and sharing it outside the workspace are decisions, not edits — they belong with the analyst. The export still runs locally for you whenever you ask.

Review happens once, at the end#

Because every edit is a saved version, you don't need to babysit each step. Let the agent work. When the session is done, ask the file for the list of changes over the last hour, and the before/after comparison for the last few of them.

Scan the changes. Look at the reasoning behind the new claims the agent introduced. Accept the run, or roll the file back to the version before the agent touched anything. The handoff is legible because the file remembers everything that happened to it. See show your work for how the reasoning side of that audit trail is captured.

Why this is different from "AI in Excel"#

Excel has no open connection like this. There's no equivalent surface an external agent can drive while the user watches and a versioned file records the result. "AI in Excel" plateaus at the cell — it can fill values, write a formula, run a macro. It cannot carry the structure of the conversation that produced those changes, and it cannot be reviewed at the level of what did the agent decide? You get a workbook that looks the same as it did yesterday, with new numbers in it, and no durable record of why.

The format is what makes the agent's work legible. The saved version history is what makes review tractable. The list of actions the agent is walled off from is what keeps the analyst's signature attached to anything that leaves the workspace.

Where the line is#

Claude reads, queries, proposes, edits. The analyst publishes, exports, ships, signs.

That's the line. It runs through two actions — publishing a change and exporting to Excel — and it's enforced by the system itself, not by trusting the agent to behave.

Coexistence, again#

Excel doesn't have that kind of open connection — but it does have you. Use both. The point isn't to replace your tooling; it's that some workflows that used to end in fog now end in a clean handoff. You can see what the agent did. You can undo what the agent did. You can put your name on what's left, and mean it.

Round-tripping a model back to Excel is the next piece of the story — what happens when you export the file the agent and you built together.

See it for yourself — open a sample .deepcell in the playground. Edit a value, watch the dependents recalculate, inspect the reasoning behind any number.