ONLINEcascade://samwarren.iov0.1.0
all projects
POTENTIATING · V2 · SCALE-SAFE REDESIGN
WEEKLY DIGEST · V1 → V2· 04 / 06

Target Account News Agent

A per-rep weekly digest, rebuilt for scale.

Every Monday, every rep wakes up to a per-account briefing on what changed in their book the week before. v1 proved the value. v2 answered everything v1 got wrong about scale — deterministic signal extraction, per-rep filtering, Slack-native delivery, observable pipelines. The upstream half of the Sales Intelligence Hub.

per-rep
DIGESTS/WEEK
7
SIGNAL SOURCES
~5 min
MONDAY READ
all answered
V1 LESSONS
PROBLEM

v1 worked. v1 didn't scale.

The first version of the per-rep weekly digest did exactly what it was supposed to — pull signal from across the GTM stack every Monday morning and hand each rep a briefing on what changed in their book the week before. It was popular. Reps opened it. Reps acted on it.

It also broke every other month. The extraction was brittle (one malformed Gong transcript could tank a full run), the ranking was global rather than per-rep, the delivery was email (so links went stale, no reactions, no threading), and the telemetry was nonexistent — if a digest was wrong, I'd find out from a rep, not from the pipeline. v1 had proven the value. v2 had to answer all of that.

APPROACH

Scale-safe from the inside out.

v2 rebuilt the pipeline from first principles. Deterministic signal extraction with typed schemas and fallback paths. Per-rep filtering driven by their actual target account list and ownership data, not a global ranker. Slack-native delivery so the digest lives where reps already work. Observable pipelines — every step logs, every failure surfaces, every run produces a manifest.

The upstream half of the Sales Intelligence Hub. The downstream half (Intel Chat) reads its output as context when a rep asks for pre-call prep on an account they saw in Monday's digest.

v1 was a proof of concept that happened to ship. v2 is the product.
WHAT V2 ANSWERS

Every lesson v1 taught, codified as a guarantee.

The rebuild turned every v1 pain into a v2 design principle:

  • ·Brittle extraction → typed signal schemas with fallback paths on every source
  • ·Global ranker → per-rep personalization driven by target account list + ownership
  • ·Email delivery → Slack-native with thread-able cards and native reactions
  • ·Silent failures → observable pipeline with per-step manifests and failure surfacing
  • ·Static digest → live context handed to Intel Chat when reps query account follow-ups
OUTCOME

Monday morning reads the same. Everything under the hood is different.

From a rep's seat, the Monday digest feels similar to v1 — same five-minute read, same per-account structure, same sense that the system understood their book. Under the hood, it's a different product: deterministic, observable, scale-safe, and engineered to be the upstream half of a larger signal chain.

The rebuild is the strongest proof of the portfolio thesis. You don't get an AI-native GTM infrastructure in one shot. You get it by shipping v1, learning what breaks, and rewriting the hard parts with the lessons baked in.

▸ stack
pythongithub actionsslackcloudflare d1openai
▸ clone it
▸ public repo · ready
The pipeline pattern: N signal sources → per-identity filter → ranked sections → Slack delivery. Three source adapters (RSS, Hacker News, GitHub releases), deterministic filter with exclude-rules, YAML identity config, GitHub Actions cron. 10 passing tests, working --demo mode with bundled fixtures.
▸ connects to
NEXT · 05 / 06
The Marketing Ops Agent
Five skills, one operating capability.