Wed 03 Jun 2026
Thirteenth pass — 2026-06-03 (Pomelli — Google's free AI marketing tool for SMBs)
One item this pass — and a different shape of item from everything above. Surfaced by Anthony 2026-06-03. Pomelli — a Google Labs × DeepMind AI marketing tool, launched 28 Oct 2025, free public beta. Verified via firecrawl_search + the official blog.google launch post. Three-step flow: (1) Business DNA — enter a URL, Pomelli scrapes the site + images and auto-extracts brand identity (tone of voice, fonts, colour palette, imagery); (2) tailored campaign ideas grounded in that DNA; (3) edit + export on-brand social/site/ad creatives, fully editable in-tool. Feb 2026 added Photoshoot (basic product photo → studio-quality shot). Most items on this radar are tools to adopt or steal. Pomelli is the first that is primarily a free incumbent commoditising a slice of Articulate's own offer — so it scores low on adoption and routes to competitive intelligence, not the install queue.
| Item | Source | P | Q | S | Total | Move |
|---|---|---|---|---|---|---|
| Pomelli — Google Labs × DeepMind AI marketing tool. URL → "Business DNA" brand extraction → on-brand social campaign generation → edit/export. Free public beta. US · Canada · Australia · New Zealand only, English only. | Anthony's forward 2026-06-03; verified via firecrawl_search + blog.google launch post |
1 | 1 | 1 | 3/9 | Skip as a tool to adopt — escalate as competitive intelligence + a positioning hook. Geo-locked out of the UAE/MENA (not available where Anthony operates), English-only, SMB self-serve, commodity-tier output, no strategy/sovereignty/consultative layer. It does not raise what Articulate ships and is not resellable surface. But it is the most strategically significant item to cross the radar in weeks: Google has put a free tool on the exact "generate on-brand SMB social content" task the MEP content layer sells. The correct response is positioning, not adoption — and the geo-gap + missing layers actually sharpen Articulate's whitespace. Flag to Articulate positioning + a Belinda/Will Stack Watch piece. |
Why P=1 (skip-tier adoption)
Could Anthony run it himself for own marketing? Marginally, and only behind a VPN — it's geo-locked to US/CA/AU/NZ, English-only, and it overlaps tooling he already runs at far higher control (the brand-voice plugin for voice, discover-brand + guideline-generation for brand extraction, the ai-image Higgsfield stack for creatives, all Belinda/Dylan-gated). It removes no hours from a real Articulate/GIG/Villa workflow that the existing stack doesn't already cover better. No daily lever.
Why Q=1
Pomelli's output is templated, SMB-grade social content — the commodity tier. It does not beat the brand-anchor + Soul Cinema + Belinda-gate pipeline on a single deliverable Anthony ships. The one genuinely interesting pattern is Business DNA: one-pass auto-extraction of tone/fonts/colour/imagery from just a URL. That's the same job as discover-brand + guideline-generation, compressed to a single automated step — worth a reference look as a front-of-funnel speed-up for the brand-discovery workflow (does Pomelli's auto-DNA beat our multi-step discovery on a real client URL? cheap to test where it's reachable). But the tool itself lifts nothing in flight. Q stays 1 on the reference value of the pattern, not the output.
Why S=1 (and why the threat axis isn't the S axis)
S measures "sellable inside an Articulate workflow." Pomelli is free, geo-blocked, and reseller-margin-less — it cannot be a billable line item. So S=1. The reason this item still matters is competitive, and threat ≠ sellability, so it does not inflate S. Three structural gaps keep Pomelli off Articulate's turf and widen the whitespace already named in the 2026-05-16 competitive scan:
- Geography. US/CA/AU/NZ only. The entire UAE/KSA mid-market Articulate targets cannot use it. (And a US-cloud Google tool scraping a regulated client's site is the opposite of the §3g data-sovereign pitch.)
- No strategy / no human consultative layer. Pomelli is self-serve asset generation. Articulate sells GTM strategy + brand + delivery under one operator — the "solo operator + AI leverage + strategy" whitespace nobody in the GCC scan owns.
- Commodity content only. It automates the cheapest layer (social creatives), not the engine (MEP), the diagnostics (Connie/IC), or the sovereign stack.
The positioning hook to steal: "Google now gives away the on-brand-social-content layer — so why does a UAE SMB still pay an agency? Because the content was never the moat. Strategy, sovereignty, and a named operator are." That's on-brand Stack Watch material and a sharpener for the MEP one-pager.
New generalisable note — the "incumbent-commoditiser" class
Sibling to the RAG-widget-SaaS skip filter (fifth pass) and the crawler/scraper skip filter (eleventh), but pointed the other way. When a hyperscaler ships a free tool that automates a slice of Articulate's own offer, it scores low on P/Q/S adoption — but it does not get filed and forgotten. It routes to competitive intelligence + positioning content. The output is a Stack Watch / MEP-positioning note and a "what's now commoditised vs what's still the moat" read, not an install decision. Adoption-skip and strategic-significance are orthogonal; this class is the case where they diverge hardest.
Class-match against failures.md and prior passes
- agnt.one (fifth pass, 1/9) — closest category match (AI marketing/content automation) but inverse handling: agnt.one was a no-name US SaaS → auto-skip-and-forget; Pomelli is Google → skip-as-tool but escalate-as-intelligence. The vendor weight is the difference.
- Corey Haines / 10-prompt pack (second + sixth passes) — "build the superior Articulate version, don't install the artefact." Same instinct, harder: here you can't even build on Pomelli (closed, geo-blocked) — the only move is to out-position it.
- Vendor-doc spot-check (ninth/twelfth passes) — applied: confirmed region lock and free-beta status against the official blog.google post rather than the hype framing ("builds your entire marketing campaign"). The "entire campaign" claim is social-creative generation, not strategy or distribution.
- No
failures.mdnegative class-match — net-new competitive-intelligence item.
Re-score triggers
- (a) Pomelli opens in UAE/MENA → re-score P (Anthony can use the Business DNA extraction as a front-of-funnel tool) and re-evaluate the competitive threat to the MEP content layer with real local availability.
- (b) Business DNA auto-extraction beats
discover-brand+guideline-generationin a head-to-head on a real client URL (where reachable) → Q promotes; consider lifting the one-pass-from-URL pattern into the discovery workflow. - (c) Google ships an API / Workspace / Ads integration that makes Pomelli embeddable or reseller-accessible → re-score S as a possible delivery component rather than a pure competitor.
- (d) A prospect names Pomelli as the reason they don't need an agency → that conversation is the trigger to ship the positioning piece; log the objection + the answer.
Where it lives (the placement decision)
Filed to canon this pass: this radar entry + a one-liner in SuperSebastian/radar.md inbox. The actionable forks are surfaced to Anthony, not assumed — both sit in other personas' lanes: (1) a Stack Watch / LinkedIn piece on Google-commoditises-SMB-content (Will + Belinda brand-fit call), and (2) a sharpening line in the MEP one-pager on what's now commoditised vs what's still the moat (Articulate positioning, Anthony's call). Sebastian proposes; he does not write to Articulate/ canon.
Fourteenth pass — 2026-06-03 (Matt Van Horn "Every Agentic Engineering Hack I Know — June 2026")
One artefact, scored as a stack (same shape as sixth pass = 10-prompt pack, eleventh pass = crawler category). Pasted by Anthony from the @mvanhorn X long-post (341k views). 22 hacks. Source is a coding-agent operator who ships OSS (last30days 27k★, Printing Press 4k★, top contributor to cpython/Go). His substrate is cmux + Codex + IDE-extension + terminal-into-Claude + plan.md files. Anthony's substrate is Cowork + the ~/ClaudeWork/ vault + Hermes Agent + Higgsfield — not a terminal-tab fleet shipping software. So most literal installs are wrong-substrate or duplicate canon. The portable value is three techniques, not tools, plus two genuine tool-gaps the article exposes.
The headline
Harvest the substrate-agnostic techniques (plan-for-the-plan · raw-Granola-to-plan · research-before-plan). Skip the IDE/terminal/coding-fleet plumbing and anything that duplicates the vault. Two real gaps surface: an explainer-video pipeline (HyperFrames) and the Granola CLI as the engine for the already-scoped Sales Call Intelligence offer (sixth pass). And read §21 — see the standing note at the end of this pass.
Scored sub-items
| Item | What it is | P | Q | S | Total | Move |
|---|---|---|---|---|---|---|
Compound Engineering — "plan for the plan" (/ce-plan → /ce-work, universal-planning mode, don't-read-the-plan) |
Plan-file forcing function; fans out parallel research, writes checkbox acceptance-criteria, then executes mechanically. Applied to deep non-code knowledge work (the strategy-doc/transcript example). | 2 | 3 | 1 | 6/9 | Act on — as technique, lifted into our planning canon, not necessarily the EveryInc plugin. We already plan-first via vault + pm:* + Rick/Kendall gates; the steal is the "plan-for-the-plan + acceptance-criteria + don't-read-the-plan" discipline as a forcing function on proposals/MEP-discovery/Connie diagnostics. Same call as Corey Haines / 10-prompt pack. |
| Granola raw-transcript → plan + Printing Press Granola CLI | Drop the raw meeting transcript (no pre-summary) into the planner; pull any past meeting as structured data via CLI. | 2 | 2 | 2 | 6/9 | Act on. Granola + the WhatsApp/calls pipeline already land transcripts in the vault (§3f). "Raw-over-summary" is a free technique change today. The Granola CLI is the engine for the Sales Call Intelligence offer (sixth pass) — pull-search-across-every-call is exactly that deliverable. |
| HyperFrames — agent-rendered explainer video | Video as HTML (script.md → kinetic-typography/captioned MP4), agent writes it, no editor/timeline. HeyGen-owned. | 2 | 2 | 2 | 6/9 | Act on. Fills a real gap — Higgsfield = character-in-scene Souls (GIG); fal/LongCat = lip-sync/long-form; nothing covers cheap captioned explainer/launch-reel video. Pilot on Articulate's own launch reel; route through Dylan/Roger + the image-critique skill for brand discipline. US SaaS but not client-data — RAG-widget/sovereignty filters don't bite for marketing assets. |
| last30days — research-before-plan | Parallel social-surface search (Reddit/X/YT/TikTok/HN/Polymarket/GitHub) before a tool/library decision. ScrapeCreators key. | 1 | 2 | 1 | 4/9 | Conditional. We have Firecrawl as ★ primary search + Exa + the eleventh-pass settled web-data layer. last30days adds social-surface breadth this radar pass should itself consume. Re-score if a parallel-social skill beats Firecrawl on a real decision. Don't add a second search substrate without proof; needs ScrapeCreators key. |
| Agent Cookie — real browser-session auth to a CLI | Hands a CLI your live browser session so it acts as you; cookie/.env sync across machines. | 1 | 1 | 1 | 3/9 | Park with trigger. Cloudflare Workers (HTTP) + Hermes skills (daemon) already own automation (§3a/§3b). The one live pain it touches is the Firecrawl-cookied-browser LinkedIn ingest auth (§3e). Re-score Agent Cookie standalone if that auth keeps breaking. |
| Proof — share plan/spec with a non-terminal reviewer | Drop a .md into Proof, send link, inline comments flow back into the loop. | 1 | 1 | 1 | 3/9 | Skip. Cowork artifacts + the per-project deliverables/ HTML site already cover client review (environment.md hard rule). Re-score if a client specifically wants inline-comment review on a living plan. |
| AgentMail — email fires a Claude session | Email an inbox → fresh session opens and works the subject/body. | 1 | 0 | 1 | 2/9 | Skip. Duplicates Hermes Agent + the WhatsApp bridge (§3a/§3f) — on email, the channel Anthony explicitly treats as dead-letter. |
| Voice-to-LLM (Monologue / Wispr Flow + gooseneck) | Speech → focused app; dictate into the agent. | 2 | 0 | 0 | 2/9 | Skip as canon — personal install only (hidden-bonus inverse: no Q/S, P-only). ElevenLabs Scribe already owns the transcription substrate (§2); WhatsApp-voice-note → Scribe → vault already covers mobile. Desktop voice-in is a personal convenience, not a stack item. |
| cmux — 4-6 parallel terminal sessions | Terminal multiplexer running many Claude/Codex sessions at once. | 0 | 0 | 1 | 1/9 | Skip — wrong substrate. We run Cowork (single rich session) + Hermes daemon + subagent fan-out for parallelism, not a terminal-tab fleet. Same generic-vs-bespoke skip as gstack (first pass, 0/9). |
No-score skips (substrate mismatch or duplicates canon — logged so they're not re-litigated)
- YOLO
dangerously-skip-permissions+ "don't read the plan" posture — the opposite of our posture for client deliverables (Rick/Kendall/Belinda gates, SEB four-line gate). Adopt the plan-discipline; reject the no-gates framing. Direct SEB-gate failure-class match. - Terminal-opens-into-Claude / sound hook — N/A under Cowork.
- Codex as second engine (Claude plans, Codex builds) — Anthony isn't shipping heavy code; the OpenAI key is debias-only (taste-pipeline step 2, §2). No second-engine need.
- Notes-as-knowledge-base (Bear / Obsidian / gbrain / supermemory) — the
~/ClaudeWork/vault is this. Same skip as Claude-Mem (first pass, 0/9). - Mosh / Tmux remote work — the sandbox→Studio Tailscale chain (§3c) is our equivalent.
- Write-your-own-skills — already canon (skill-creator + the persona/skill stack + Taste→SkillOpt lifecycle). Affirms, adds nothing.
- OS contribution + paying $1-3 to subscribe on X for PR notifications — distribution/relationship play, tangential to Articulate.
Class-match against failures.md + prior radar
- Corey Haines (second pass) + 10-prompt pack (sixth pass) — the dominant pattern: lift the technique, don't install the viral artefact. Applies to all three act-ons (CE plan-for-the-plan, raw-Granola, HyperFrames-as-input).
- SEB-gate four-strike (2026-05-24) — Van Horn's YOLO + "don't read the plan" is exactly the vibe-mode/no-gate framing that fires our worst failures. Adopt the structure (plan → acceptance criteria → execute → verify), reject the posture.
- Claude-Mem 0/9 + gstack 0/9 (first pass) — notes-as-KB and parallel-session-orchestration are the same generic-vs-bespoke / duplicates-vault skips.
- RAG-widget skip filter (fifth pass) — not triggered; no chatbot/widget items in the article.
Re-score triggers
- (a) CE plan-for-the-plan technique lifted into MEP-discovery / Connie / proposal canon and runs once on real client work with measurable laziness-reduction → P→3, total 7/9.
- (b) Granola CLI wired as the Sales Call Intelligence offer engine with first billed Gong/Granola audit → S→3, total 7/9 (shares the sixth-pass Sales-Call-Intelligence trigger).
- (c) HyperFrames pilot ships a launch reel faster/better than current (none today) → P→3; scoped as an MEP/GIG explainer-video line item → confirms S, total 7/9.
- (d) last30days-shaped parallel-social skill beats Firecrawl on a real tool decision → P→2, total 5/9, conditional-install.
- (e) Firecrawl-LinkedIn cookied-browser auth keeps failing → re-score Agent Cookie standalone.
Standing note — §21 "AI Psychosis" (not scoreable; surfaced to Anthony)
The author's most honest section: agents were meant to do the work, instead every operator he knows is working the hardest of their life — "the greatest video game ever made," build-addiction, launches with no users, losing the people around them. Anthony's profile matches the risk shape — solo operator, ~10 active surfaces (Articulate · GIG · Roche · BossCouple · Villa · MediaServer · the persona stack), intensity-driven. Sebastian's flag, not a score: the discipline that protects the business here is fewer surfaces shipped harder (the operator-scan lesson, line ~273) and "build something people want, even if 'people' is just you." Worth a Mother/PM conversation, not a stack move.
Filed to canon
This pass + a one-liner to SuperSebastian/radar.md inbox. Act-on forks sit in other lanes and are proposed, not written (Sebastian's lane rule): (1) lift the plan-for-the-plan discipline into Articulate/offer.md MEP-discovery + Connie's playbook (Anthony/Connie call); (2) Granola CLI as the Sales Call Intelligence offer engine (sixth-pass offer scaffold, Anthony's call); (3) HyperFrames pilot in Dylan/Roger's lane, Belinda-gated before any external use.