Working with curated data doing complex work. We design automations to run our system. Here are a few.
8 automations · grouped by usePulls a recorded meeting transcript, classifies the call type, and extracts twenty-five sourced data points — every claim traces back to an exact quote.
Produces a structured HTML debrief with action items, commitments, fear and greed signals, and a strategic frame that names where to play and how to win.
Starts a billable clock for a named client and loads that client's full knowledge base — intelligence, voice, and design layers — into the session.
Makes every subsequent draft, decision, and action client-aware AND time-tracked, with the clock surviving session resets and two-hour focus blocks.
Pulls billable hours for a client over a date range, fills missing notes interactively, and renders an invoice in the firm's native format.
Ships a sendable HTML and PDF invoice with a sequential per-client number — the bookkeeping step that used to stall.
Loads the canonical voice rules, parses the request for content type and topic, optionally sources from an existing debrief, and writes.
Produces LinkedIn posts and website copy in the firm's voice — one insight, fully expressed, no brochure language, no engagement bait.
Walks an internal process tree with the operator, layers any primed client voice on top of global rules, drafts, and scores against a clarity rubric.
Writes and refines communications in the operator's exact voice — emails, posts, bios, docs, replies, anything professional.
Pulls a Granola transcript or fetches a YouTube video's captions, asks what the source should help with, and extracts through five lenses — every claim cited to a transcript quote.
Produces a tagged markdown file of atomic learnings plus a synthesized HTML idea sheet — same five lenses across calls and videos, so the directory becomes a grep-able commonplace book that compounds over time.
Interviews the operator one question at a time, walking down the decision tree of a plan or design in dependency order — answers what the codebase can answer, asks what only a human can answer.
Stress-tests a plan until both sides share an unambiguous picture of what will be built and why. No hand-waving accepted.
Fans out multiple background agents — each with a per-task state file on disk — so one operator can orchestrate several unrelated tasks in parallel. The status dashboard updates only on change; a FIFO blocker queue surfaces failures one at a time.
Runs three to five unrelated tasks at once across different repos, with each agent checkpointing to disk so the work resumes from a fresh session after a /clear or crash.