01 Ship software
From a prompt to a running product.
Attach a brief and it writes the code, starts a dev server, then screenshots and records the result to show you. Click any element in the live preview, say what to change, and it ships the fix.
Hand it a goal. It reads your files, digs through the web, runs the code, and puts the finished thing on your desk — a dashboard, a brief, a model. The file itself, not a description of it.
Real runs, captured end to end — it takes your context, does the work, and hands back the finished file. Software, spreadsheets, research; the same system does all of it.
Attach a brief and it writes the code, starts a dev server, then screenshots and records the result to show you. Click any element in the live preview, say what to change, and it ships the fix.
Point it at your growth model. It charts the trend, builds a live revenue dashboard, and exports a board-ready PDF — graphs, tables, and a written summary, all attached.
Give it a question. It browses, extracts, sizes the market, and charts what it finds — then hands back a polished briefing with the numbers and the source list, as a doc and a PDF.
Identify axes: price, API, ecosystem, fit for SMB.
3 browsers: Salesforce, HubSpot, Zoho. Concurrent.
Walking pricing pages, docs, G2. 186 URLs visited.
Normalize tiers, de-dupe features, score fit.
Memo with evidence references. Ready to forward.
memo.md · comparison.csv · pricing.xlsx · sources.json
Every item here is a real run, with real artifacts attached. Fork any of them as a starting template for your own work.
Runs web research in parallel. Extracts pricing, features, reviews. Returns a comparison table and recommendation memo.
Splits into finance, market, legal, ops subagents. Reads uploaded Drive docs. Ships a master report, risk register, and exec summary.
Reads .xlsx files, profiles data quality, flags anomalies and outliers. Returns a cleaned workbook, dashboard image, and findings report.
Pulls vendor, amount, due date, line items from invoice PDFs. Normalizes to CSV for accounting. Surfaces PO mismatches.
Parses contracts. Highlights renewal, termination, indemnity, liability. Returns a red-flag summary and follow-up checklist.
Ingests mixed docs, slides, spreadsheets. Synthesizes into a board-style summary with decisions, recommendations, appendix.
Reads the RFP plus internal source docs. Drafts section-by-section with evidence references. Returns submission-ready drafts.
Extracts rent roll and operating statements. Builds a 10-year cash flow with sensitivity analysis. Returns the full model.
Summarizes transcripts. Extracts action items, owners, due dates. Drafts the follow-up email and task list.
Clones the repo, maps the architecture, proposes a targeted debugging plan, suggests fixes with a test strategy.
Agents that touch real documents and real systems need real controls. Here is what that looks like.
Runs execute in a private sandbox per account. Nothing is used to train anyone's model. Delete a run and it's gone.
Plans, tool calls, file reads, URLs visited — all logged and exportable. Audit a run like you'd audit a junior.
Per-run and per-day budgets. Agents ask before crossing a ceiling. No surprise bills, no infinite loops.
Connect Drive, GitHub, or your data warehouse with read-only keys. Revoke from the dashboard at any time.
Flat monthly price. Fair-use runtime. No per-seat upsell until you genuinely have a team.