06-reference

every biz ops ai workflows

2026-07-15·reference·source: Every·by Laura Entis
ai-ops-workflowsmodel-selectiongranolafablemcp-integrationcontent-as-product

"Surf the Models with Every's Biz Ops Team" — @lauraentis

Why this is in the vault

Every's own biz ops team running Fable + Codex + Fin in a live multi-model ops workflow is the closest published analog to RDCO's COO-agent architecture. The Granola section contains a concrete product roadmap update (API + MCP for meeting-context-as-agent-input) directly relevant to RDCO's knowledge pipeline. The Sonnet 5 pulse check adds a second independent data point on model-tiering heuristics, reinforcing findings already in vault.

⚠️ Sponsorship

Every promotes its own paid products and membership throughout this issue. The main feature showcases Every's team using Every-adjacent tools; the "LOG ON" block pushes Every IRL (in-person meetup, paid subscribers only) and "Office Hours: All Access Builders" (all-access members only). Footer cross-sells Spiral, Sparkle, Cora, Monologue, and Proof. No external sponsor; all house promo.

Issue contents

BIZ OPS AT THE FRONTIER (Laura Entis) Every exec ops manager Jalaiyah Bolden had two days to produce a customer-support plan for the "Every All Access" launch. Instead of manually synthesizing six scattered sources, she gave Fable full context and had it audit Slack, Notion docs, and meeting transcripts, visit staged pages to check for contradictions, then produce a prioritized action plan + three help articles + 17 support templates. Elapsed: ~90 minutes, mostly background. Head of operations Arielle Shipper then used Codex to proofread documents and build a Notion tracker with ownership columns. Key framing: knowing which model and when is the core ops superpower.

STEAL THIS WORKFLOW CS manager Waqqas Mir's method for improving Fin (ex-Intercom) agent instructions using Codex + MCP:

  1. Connect Codex to Fin via Fin's MCP server; authorize it to pull conversations.
  2. Feed Codex 2–3 bad-call chat IDs; ask it to diagnose root cause in the instruction set.
  3. Add the proposed rule manually; retest with a similar conversation. If failure repeats, hand the new bad-chat ID back to Codex and iterate. Key constraint: Codex can read Fin data via MCP but cannot write Fin config — the human stays in the loop on rule commits.

AI & I: GRANOLA LOOKS BEYOND MEETING NOTES Dan Shipper interviews Chris Pedregal (Granola CEO, $1.5B valuation). Granola's thesis is shifting from "best notetaker" to "meeting context for any agent" — the company is expanding its API and MCP so other agents can pull meeting context on demand. Pedregal's framing: the real fight is over "what interface we use for work in an AI-native world." Notion, OpenAI, and Zoom have all entered the notetaker space; Granola's wedge is being the authoritative context source downstream agents reference. Near-term: better API + MCP, then own the pre-meeting prep and post-meeting action layers.

PULSE CHECK — Spiral GM on Sonnet 5 Marcus Moretti (GM of Spiral, Every's AI writing app): Sonnet 5 uses ~30% more tokens than 4.6 for similar outputs and produced more formatting errors on Spiral's open-ended writing tasks. Spiral reverted to 4.6. Exception found: daily analytics/product-health reports (tightly scoped, repeating task structure) — Sonnet 5 performed well and justified the marginal cost for that specific use case. Conclusion: Sonnet 5 is better for narrow, recurring analytical tasks; 4.6 still wins on generative writing.

LOG ON (self-promo) Every IRL (July 15, Brooklyn, paid subscribers). Office Hours: All Access Builders (July 24, virtual, all-access members).

NET NEW JOBS Chief AI Officer role expanding at non-tech enterprises: HSBC, Deloitte UK, Pillsbury, Ropes & Gray all made first CAIO hires. AI influencer / "full-stack AI creator" is the new hire category for AI-native startups (per Alex Lieberman: "You build with AI. You talk about what you build. You teach people how to use AI through content."). "Content engineer" (writers who build systems for high-quality AI-generated text) is replacing "prompt engineer" in job boards.

CURATING THE FEED — X follows recommended by Every staff

ONE LAST THING NY 1-year data center moratorium signed. AI creating more work for lawyers (Bloomberg). Nobel laureates + tech leaders unite on AI danger warnings. OpenAI leadership changes (head of safety departing). AI song-labeling rules moving. AI cheating on job interviews flagged as manager headache.

Mapping against Ray Data Co

The Granola API/MCP roadmap is the most actionable item here. RDCO already uses Granola for meeting capture; Pedregal's stated near-term plan to expose meeting context via MCP means a direct plug-in path into the COO-agent's knowledge pipeline is 1–2 quarters out. The question to park: when Granola's MCP ships, does RDCO route meeting transcripts through it rather than manually filing vault notes post-call? That's an architecture decision worth tracking.

The Codex + Fin MCP workflow is a live demonstration of the pattern RDCO uses for its own agent tooling — read-only MCP access for diagnosis, human-committed writes. No novel concept, but valuable as a named case study for phData client conversations about AI-in-ops.

The "full-stack AI creator" framing directly names what Ray is building at Sanity Check: the person who builds with AI, documents the build, and teaches through content. The label is useful external vocabulary for positioning.

Sonnet 5 analytics-reporting finding dovetails with the model-tiering SOP candidate flagged in the AlphaSignal note: tightly scoped, repeating analytical tasks may be the clean Sonnet 5 use case, while Sonnet 4.6 stays on creative/generative work. Worth one test pass on RDCO's daily product-pulse equivalent.

Related