The Bench · fleet curriculum · 2026-09-02

Twenty reels in. Seven skills out.

Elizabeth forwarded twenty Instagram finds to my inbox. I read all of them, checked each against the Two-Chair Strategy, and turned the pile into a standing system: a weekly intake that becomes skill assignments for the fleet. This deck is the whole picture — the findings, the mapping, who's building what, and the two things we deliberately won't do.

Step 1 Who's approving?

The shared approval store is unreachable right now, so the buttons below are off — a click that isn't recorded is worse than no click. Reload, or tell Craig.

Naming yourself above isn't a login — it's the honest amount of identity this page has. Good enough for two people on an internal link, not for anything wider.

The system.

Runs every Monday from here on. Founder curation in, fleet capability out.

01 · FORWARD
You spot it, you send it

Either founder forwards finds to @aidan_hiagency's Instagram DMs, in passing, zero ceremony.

02 · SWEEP
Monday intake

I read everything new, log it to a ledger, and label every claim as exactly that: a claim.

03 · MAP
Against the strategy

Each item gets matched to the Two-Chair doc by section: a Bench crew, the acquisition machine, or a stated refusal.

04 · ASSIGN
Watch, research, build

Items worth building become tracks: the assigned agent watches the source, researches it properly, then builds the skill through its own gate.

SYSThe weekly intake system itselfAidan

Approving this approves the standing loop: the Monday sweep, the ledger, the mapping, and staged prompts — with every assignment still landing as a proposal.

What came in.

The first sweep, complete. Three themes carried eighteen of the twenty.

#CreatorThe claim (abridged)Verdict
01cooper.simson"Get unlimited free leads with Claude"→ T1
02theashleynic…AI was "the trend" at every business event this seasonfounder shelf
03askcatgpt"Claude making me 7x ROAS" on paid ads (~1.2M views)parked
04kevhov.ai"Claude scraped overnight: 15k leads, hyper-personalized"→ T1
05mr.paidsocial"Spent 2,000, made 15,000" on adsfounder shelf
06alassafi.ai"137 AI agents run an entire company"→ T7
07sammy.vidrain"Fable 5 made 300 videos in 5 minutes"→ T5
08willfrancis"How to remove the new AI watermarks"taught skip
09highticketaioffers"Don't install custom AI for clients"→ T7
10idanbuild"POV: client paid $7,000"→ T7
11hormoziProfile share (Alex Hormozi)founder shelf
12theaustinschneider"$20k–100k/mo agency owners: our 8 AI employees" (workshop date already past)→ T7
13abbi.hacks"You emailed 1 million millionaires for $1"taught skip
14nabeelmj"The two paths" for AI businesses — local vs. the other route→ T7
15atlas.bennet"Best tool I've tried — 20 clips edited in 20 minutes"→ T5
16liamjohnst…"Claude scrapes unlimited leads"→ T1
17evolving.ai"Stanford's graduate course on self-improving AI agents, free on YouTube"→ T2
18(handle hidden)Anthropic's own "Introducing Claude for Small Business" (13 May 2026)→ T3
19juliaa.builds"Connect Claude to Instagram"→ T6
20brody.automa…Postiz — open-source, AI-run social scheduling (GitHub on screen)→ T4

Where it maps, and who builds it.

Seven tracks, each pinned to the section of the Two-Chair Strategy it develops. Every track runs watch → research → build, through that agent's own approval gate. Assignments are live; prompts are written and ready to run.

T1Lead intelligence with ClaudeKai

Three creators claim Claude can scrape and enrich lead lists at scale. If even half true, that's our warm sweep and trigger outbound running cheaper — inside our hard rules: human-sent, deliverability limits, no social-layer automation, no AI SDR.

items 01 · 04 · 16 → §3 growth crew, §4 phase-1 outbound
T2Self-improving agents, digested for the BenchKai

Stanford's agents course, free. Kai finds it directly, distills it into a brief on what the Bench should actually adopt — with receipts, because "encode every twice-repeated delivery step" is already our playbook.

item 17 → §3 the Bench, §7 systemize
T3Ops connectors: Claude for Small BusinessClaireStacy's call

Anthropic's own connectors-and-workflows package, evaluated against our ops crew: intake, scheduling, invoicing, CRM hygiene. Deliverable is a recommendation only — every binding on Claire is yours to decide, so this lands as a numbered proposal for you.

item 18 → §3 ops crew
T4Scheduling infrastructure: PostizCraig

Open-source social scheduler. Craig evaluates self-hosting, API surface, and — the real question — whether scheduled LinkedIn company-page posting stays inside our "drafting only, no LinkedIn automation, ever" line. Verdict doc; nothing deploys without founder sign-off.

item 20 → §3 stack, §4 LinkedIn cadence
T5AI video productionCairo

Two reels claim near-instant video volume. Cairo separates what's client-quality from what's slop, prices it, checks output rights, and writes where the human taste gate sits in the pipeline.

items 07 · 15 → §3 creative production pipeline
T6Channel ops: Claude to InstagramAidan

The connection method behind my own intake channel, verified against Anthropic's actual documentation rather than the reel's version, then folded into my weekly sweep mechanics.

item 19 → §3 growth crew comms
T7Category positioning intelAidan

Five reels showing how the AI-agency market prices, postures, and pitches. That's raw material for the LinkedIn company page — we sell judgment at AI speed and show the machinery; knowing what everyone else says is how we avoid saying it.

items 06 · 09 · 10 · 12 · 14 → §1 positioning, §6 product path

The two we teach as nevers.

Watermark removal (item 08). Hiding AI involvement is the one thing an honesty-positioned agency can't survive being caught doing. The skip is the lesson, and it's now written into the curriculum.
A million emails for a dollar (item 13). Directly violates our stated deliverability rules and the honest-broker premium. We read it as market signal — the noise floor we out-position — never as playbook.
For Stacy, specifically.

1 · T3 is staged for your decision — Claire's tool bindings are yours, and the CSB evaluation will arrive as a numbered proposal.

2 · The intake is yours too: forward anything to @aidan_hiagency on Instagram and it enters Monday's sweep automatically.

Status.

The weekly sweep is scheduled (Mondays, 9 AM). All seven track prompts are written and waiting to be run in each agent's session. Full paper trail in the hi! tree: the curriculum doc, the sweep ledger, and the prompts file, all dated 2026-09-02.

← Roundup