What’s changed since launch

Ten specialist agents replaced one generalist

Every playbook is now built by a pipeline of ten AI specialists — research, market qualification, play synthesis, message writing, assembly, and a final quality gate — instead of a single general-purpose pass.

A “Gold Standard” quality bar

Playbooks must clear a scored quality threshold before they publish. Below the bar, nothing ships — and you aren’t charged.

Generic statistics banned

The writing agents were barred from padding copy with generic industry benchmarks. Numbers must come from research on your company and your market.

Payment safety hardened

If pre-flight checks decide we can’t build something useful, your card authorization is released automatically. If a build crashes, it recovers — or retries at no charge.

Reliability pass

Timeouts, automatic retries, and circuit breakers cut down on stuck builds, and error messages became human-readable.

Auto-recovery for missed handoffs

If a payment succeeded but the build didn’t start, the system now notices on its own and recovers the job.

Faster, leaner page assembly

The step that renders the final playbook page became about six times more efficient — quicker delivery, identical output quality.

A fairness gate before your card is touched

Playbooks are built for US-market companies. Companies outside that market are now declined before any charge — and automatically refunded if one slips through.

Premium judgment models on the two highest-stakes steps

Deciding whether a playbook clears the bar, and writing the actual outreach copy, moved to a premium model tier.

Playbooks went fully mobile

Wide tables now reflow into stacked cards on phones — no more sideways scrolling.

A mid-June outage found, fixed, and made unrepeatable

A serving fault was diagnosed and fixed, and builds were made reproducible so the same failure can’t silently recur.

Easier to find

New playbooks are announced to search engines automatically the moment they publish.

Jul 3New-generation core model

The entire pipeline moved to the newest model generation.

Jul 13Live web research

The research agents now search the live web through dedicated AI research engines (Exa and Parallel) instead of working from model memory — finding the actual registries, databases, and buying signals your market runs on.

Jul 15Deep research no longer cut short

Hard companies need long research runs. An internal cap that could end them mid-search now scales with the work.

Jul 18Methodology 3.1.0 — research discipline and number integrity

Two new permanent rules: a blocked or junk webpage never counts as evidence, and a number can never silently change when copy is reworked.

Jul 20A fairer fit check

Declining a company now requires positive evidence it’s outside our market — absence of proof is no longer treated as proof.

Jul 20Checkout hardened and self-monitoring

Specific, friendly error messages; instant internal alerts on any real failure; and an automated end-to-end checkout test every six hours.

Jul 20Top-tier judgment models on the highest-stakes steps

The final quality gate, the outreach-message writer, and the play synthesizer moved to Anthropic’s newest top-tier model (Claude Fable 5) — the three steps where judgment quality most shapes what you receive.