Agent workflow
Output formats demo
Produces a website, PDF, Word doc, and spreadsheet.
Claude
Drop in a meeting transcript; out comes a one-page brief: decisions, owners, follow-ups, and the questions still open. Strict structure, no "key takeaways" filler.
A prompt that produces a Runway shot designed to match-cut into a real plate — same lens, same motion, same colour temperature. Stop fighting the seam.
A single Veo prompt that produces 5–8 seconds of B-roll that cuts. Specifies lens, camera move, lighting, grade, and motion arc — no overcooked anamorphic. Battle-tested on a documentary cut.
Two-step chain: extract the meaningful user-facing changes from a diff or PR summary, then write release notes in the voice you actually ship in. Drops refactors, dep bumps, and test-only churn.
Hand it 5–20 articles or PDFs; it returns a synthesis with claims, counter-claims, and citations — and refuses to invent any. Built around a strict tool-call loop with a citation-check pass.
A single Claude prompt that does a real code review — flags correctness bugs, hidden race conditions, and missing null-checks, and refuses to comment on style. Tuned over hundreds of real PRs.
Studio-grade product packshots for e-commerce. Specifies sweep, key+fill ratio, micro-reflections, and the colour reading that keeps your hero image consistent across SKUs.
An agent workflow that watches a Slack thread, classifies it (bug / feature / question / kudos / noise), files a Linear issue if it's a bug, and replies in-thread with the ticket link. Three tools, four steps, branching.
One prompt, one paragraph in, a positioning brief out: who it's for, what it replaces, the one promise on the homepage, and three hero-section headline drafts. Zero hedging.
Three-step chain that turns a one-line subject into an editorial portrait that looks like it belongs in a real magazine — not stock-photo Midjourney. Briefs in Claude, renders in Midjourney v7, then critiques and rewrites.