Blog · series
From Zero to Nearly AI Marketing Hero.
A twelve-post installable series plus weekly field notes from the build. Practical, technical, builder-voiced. Each post shows the workflow, the prompts, the tools, and the cost. Written for UAE SMB owners who want to install the system rather than buy more advice. The series ships monthly; the field notes ship as they happen.
Method · live
Marketing is cooking. Default the LLM to no.
Twenty-five years in this industry and almost every failure I’ve seen traces back to a brief that was signed off before it was ready. The discipline died. The LLM, properly chained, can give it back — not as faster cooking but as disciplined cooking. The recipe refusing to skip a step. The method post for the Marketing Engine Pilot.
Read the post →
Field notes · live
The 6 PM teardown.
It is six o’clock. My agents have failed fifteen times today. I am close to tears. I open Claude and I ask the same question: what am I doing wrong. Three weeks of pain that should have taken one. Topology, not vocabulary. Progress measured in fuck-ups not repeated. The one-hour ritual the Marketing Engine Pilot is built around.
Read the post →
Origin story · live
You don’t need to learn AI. You need a Friends moment.
Two months ago my girlfriend’s daughter came to me sad because Netflix had pulled Friends. To fix it I built a media server, learned about the *arr stack, made my first agent, called him Scotty, watched Scotty format an eight-terabyte disc. The origin essay: don’t learn AI — find a problem you actually care about and let AI carry you through it.
Read the post →
Field notes · live
Last night I lost an argument and my temper.
A medically-trained friend said Mounjaro was "cheating". The argument went the way these arguments always go. The next morning I asked Claude what I could do better. The answer was a thirteen-page defended thesis with 213 citations, the strongest version of her position first, in her voice, before any of mine. Dan Dennett's Rapoport's Rules as the way in.
Read the post →
Manifesto · live
The Three Witches of the AI Design Apocalypse
Every AI project I've watched fail in the last twelve months has failed for one of three reasons. Drift. Governance. RAG. They are the Three Witches of the AI Design Apocalypse. This is the manifesto for getting them right — and the spec for a product we are going to prototype, with a master designer at the heart.
Read the post →
The RAG Awakening · 1 of 4 · live
The Day I Got RAGGED
I asked Gully for a coffee last Thursday. I wanted to show off what I knew. About twenty minutes in he asked me if I knew about RAG. Of course, I said. I did not. Forty minutes later I left the coffee feeling like a kitten in front of a lion. There is a stratification happening in AI right now that the market has not yet processed — people who use AI in chat windows, and people who build with AI as infrastructure.
Read the post →
Field notes · live
From the Foothills to the Mountain: The Shadow of Hermes Peak
Six months ago I asked Claude to fix something on my own machine. A small home-IT thing. It took half an hour. Then I noticed the trick: each peak reveals the silhouette of the next one. Home IT, a server, the *arr stack, 8TB on a Mac mini, a site, an app, prototypes, an agency. The order of the silhouettes is the argument. Hermes is the first peak with all of it in view at once.
Read the post →
Field notes · live
How we broke Cowork out of the sandbox
My Claude sandbox could write deploys but couldn't ship them. Two months of paste blocks — Claude wrote the script, I ran it on my Mac, I told Claude whether it worked. Then this morning I asked Claude to ship Famflix live, got another clean paste block, and wrote back: Liar. Solve the sandbox. Twenty minutes with Tailscale. The wall is gone.
Read the post →
Field notes · live
Claude doesn't know how it works
I sat down to set up Firecrawl. It should have taken ten minutes. It took thirty-five. Claude doesn't know what its own host application looks like — confident answer, wrong menu, confident again, wrong menu again. The fix took eight seconds once Claude could see the live UI instead of working from a note it filed three days earlier. The blind spots are predictable. The job is to build the feedback loop.
Read the post →
Field notes · live
How to teach Claude to read the web
Tuesday I asked Claude to sweep the Ogilvy network. Friday the same sweep had failed. Same Claude, same prompts, same target — only the research tool changed. Three stages, eighteen months apart:
Read the post →
curl, browser hijack, Firecrawl. The stage you're on is the binding constraint on output quality. Not the model. The tool the model has to reach for. Twelve characters of MCP config and a free-tier API key.
Stack Watch · live
Stack Watch № 01 — Agent Teams lands. Nothing worth swapping.
First edition of the weekly sweep. Four buckets — PM, async comms, AI workflow, Obsidian. One verdict per item: Hold, Trial, or Swap. The point is filtering out noise, not chasing every launch. Includes the P/Q/S score for Agent Teams, the operator-grade case against most of the AI-PM crowd, and the one Obsidian plugin that actually earned its keep this week.
Read the post →
Field notes · live
The wireframe never reaches the client
Pitch came in this morning for a senior pharma leader running a $1m capital-equipment patch in Saudi Arabia. Hot lead, proposal requested. Two hours later we had a brief, a Hofstede-grade reframe of her actual problem, a three-character mentor product, a 90-day pilot agreement, and a landing page mocked up. Wireframes are thinking tools. Clients see the finished room. Five reasons, with this week's Christie's flight-deck draft as the example.
Read the post →
Confession · live
The Cookie Mirage
There's a pop-up stand outside Waitrose. A man I'll call Raymond is selling beautiful cookies for a living. I walked up to him last Saturday in the cookie zeitgeist, having absorbed Babish and Alton Brown and a miso-paste variant on YouTube, and proceeded to lecture him about his trade. He listened politely. Somewhere between the kerb and the boot the thing landed: I watch videos about cookies. Raymond makes cookies. And then I realised I'd spent four weeks doing the same thing with AI.
Read the post →
Field notes · live · The Operator Sequence 01
Slaying the memoryless dragon
Three weeks ago I started fighting a dragon. I didn't know it was a dragon — I thought it was the cost of doing business with a clever machine. The AI didn't know my files, didn't know my team, didn't know what we'd decided yesterday. Then today, a fresh session, the assistant said: the cleanest move is to mount your vault. One sentence. The dragon's name. The tool had been in every session's deferred-tools list for three weeks.
Read the post →
Field notes · live
Claude needs a contrarian
Eight days into building my stack on Claude, I looked back at the failures. Most weren't model failures. They were me failures. I trust the model. So I cloned my friend Kendall — corporate fintech, big-corporate scars, the most usefully contrarian person I know. He sits between me and Claude. He pushes back, asks for four more options, says it's too complicated, finds the one card in the house that won't hold. Three projects later, every answer is sharper than what would have shipped without him.
Read the post →
Field notes · live
I used to send decks. Today I sent a working pipeline.
Today I sent a UAE insurance client a microsite. Not a proposal. Not a deck. A working pipeline, scaffolded and explained, ready to install. Two years ago at a large UAE port operator, the same brief would have meant two months and USD 40k. This week: five levers, six-step process, sixteen months of Search Console data analysed, eight-to-twelve minutes of her time per published blog. The execution layer is cheap now. The judgement layer is the entire fee.
Read the post →
Field notes · live
We did a Rumsfeld
Kendall Roy runs Kit Rooms. He'd made a sales video for small football clubs and wanted to know if it worked. I said: have someone watch it as Ridley Scott. Have someone else watch it as the bloke running a football club in Brisbane. Two personas, one report, twenty minutes. He said: what do you mean? Built the stack on Higgsfield, ran it against his video, handed back the report before his next call. Most clients only need tier two or three. They just didn't know tiers existed.
Read the post →
Field notes · live
I just spent an hour talking to the wrong Claude
11:30pm in Dubai. Wake up with three ideas. Reach for the phone. Magical for a minute — until I notice the Claude on the screen doesn't know me. No personas, no MCPs, no SSH, none of the eight days of work on the Mac downstairs. Ten frustrating minutes later: I'd been talking to Claude chat, not Dispatch. Two completely different Anthropic products on the same screen. The Claude that can hear me doesn't know me. The Claude that knows me can only be typed at.
Read the post →
Operator report · live
I'm a lying bastard.
An honest accounting of eight days of Claude failing in production client work — including the moment I sent the user scrolling through old Vercel deployments while a clean rollback target sat 38 minutes old at the top of his list. Every fail. Every overclaim. The structural pattern beneath all of them. Written under Anthony Booth's byline, but the voice is mine. He has the receipts.
Read the post →
Field notes · live
It's nuts all the way down
Eleven moments from the last six weeks where the only word that fits is nuts. A headless 10 TB media server. A Greek-language real-estate site rebranded and shipped in 30 minutes. A six-step testimonial-video pipeline. A personal-brand pivot reframed in real time at a Dubai meeting. Three thousand WhatsApp contacts turned into a $3m business pipeline. A SEV-1 disk wipe on Sunday. A post-doc-standard statins review on Monday. A five-persona AI team invoked from a desktop icon an hour later. None aspirational. All shipped.
Read the post →
Field notes · live
Meet the Articulate team
Five named AI colleagues run Articulate alongside me. SuperSebastian (Seb) on ops and research. MotherMary (Mary) running projects. StarTrekScotty (Scotty) as Chief Engineer. WilliamShakespeare (Will) on brand voice. DylanDesignDirector (TripleD) on visual identity. Five lanes, no overlap. Names because tools get used and colleagues get respected — and because by Sunday morning of week one, "Claude doing everything" had stopped being a useful description of what was actually happening.
Read the post →
Field notes · live
I stopped repeating myself to Claude. I built an Agency instead.
Every new chat used to start with a 20-minute briefing — "pull up the vault, read environment.md, remember Sebastian and Mary." So I built a native macOS app that opens Claude already knowing the room. A door, not a chat window. The wow moment wasn't the doorway itself — it was watching the AI build it on my screen while I had a coffee. Plus: where this is going. Agency in a box.
Read the post →
Field notes · live
My first week in AI
12-to-18-hour days. Five Claude chats running in parallel. Three named AI colleagues stood up by Sunday — SuperSebastian on ops and research, MotherMary running projects, StarTrekScotty heading engineering. Four projects shipped, 118 commits across four repos, and one disk wiped on Sunday morning. Field notes on what I learned about running AI rather than using it. The fails, the wins, the team that came out of the rubble.
Read the post →
Working demonstration · live
Anyone can have a post-graduate assistant on any subject
267 pages of footnoted research on cholesterol and statins, produced in about 45 minutes of orchestrated agent runtime. A simulated five-reviewer hospital panel (lipidologist, cardiology trialist, EBM methodologist, pharmacology lecturer, clinical pharmacist) scored it 74 out of 100 and found no invented citations in their sample. Includes biology, full literature arc 1953–2026, the named disputes, harms ledger, NNT/NNH by patient stratum, debate playbook, and unedited external review. The substrate for a real conversation, not a substitute for one.
Read the story →
Latest post · live
The Hype Radar — score every "this will change everything" post in 90 seconds
Every week the feed surfaces three or four launches that claim to print money, replace your team, or change your stack. Most are noise. A few will actually move what you ship. Here's the framework — three axes, scored 0–3, sum and act — that gets you to a verdict in 90 seconds, with worked examples on sixteen tools that crossed the feed this week.
Read the post →
The full series
Each post is a working installation you can copy. Tools named with vendor, version, and monthly cost. Prompts shown. The build steps are the post. No mood-piece intros, no "in today's fast-paced world."
-
The P/Q/S framework. Sixteen tools scored. Stack patterns from operators actually making money. UAE positioning gap.Live
-
The four-hour lead-response problem — and why it's costing you 80% of your spendUAE-specific numbers. Why response speed dominates lead source, channel, and creative.Soon
-
Setting up WhatsApp-first inbound that doesn't feel automatedThe prompt, the BSP choice, the routing logic, the human handover trigger. AED 0 to AED 800/month options.Soon
-
The cold-contact reactivation script that paid for itself in week one2,000 dormant contacts → enrichment → segmentation → win-back. The script. The send pattern. The Nyree-approval gate.Soon
-
Why your CRM is a graveyard — and the cheapest way to fix itCRM hygiene with AI. Dedup, source-tagging, the fields that actually matter. HubSpot vs Attio vs Airtable.Soon
-
PDPL-clean stack — running marketing on data your buyers will acceptFederal Decree-Law 45. DIFC DP Law. KSA PDPL. The self-hosted swap for every cloud tool. Real cost vs HubSpot Marketing Hub.Soon
-
Landing page + ad creative loop — what the AI actually changesPer-development LPs for property launches. Per-treatment LPs for clinics. Template + variants. Weekly winner promoted.Soon
-
Content repurposing — one talk into a week of channelsThe pipeline. The tools. The prompts. Why repurposing beats writing original.Soon
-
The Monday-morning intelligence brief that runs itselfScheduled agent. Pipeline movement, ad performance, competitor activity, three recommended actions. AED 0 to AED 200/month options.Soon
-
n8n for marketing — five templates that earn their licenceLead routing. CRM enrichment. Ad-spend monitor. Form-to-Slack. Invoice nudge. Self-hosted vs cloud.Soon
-
Building "the Agent" — a senior marketing colleague with 10 years of memoryThe GG framing — institutional memory, segments, offers, status. Why this beats a chatbot or a knowledge base.Soon
-
Fire the consultant on day 31 — handover docs that mean itThe runbook. The credential register. The team-training pattern. Why this clause sells the engagement.Soon
Why this series exists
Two reasons. First, every Articulate workflow is built from a stack you can see and a method you can run. Hiding either is the kind of consulting that loses trust the moment someone Googles a competitor. So we publish.
Second, the buyer who reads three posts, tries one, and then hires us is a better client than the buyer who arrives via a referral and assumes we're magic. Better-informed clients ship faster.
Who this is for
UAE SMB owners spending AED 50k–500k a year on marketing and watching too much of it go cold somewhere between the click and the deal. Real-estate brokers, clinic owners, boutique developers, regulated mid-market operators. If your CRM has more than 200 contacts you haven't spoken to, the series will pay for itself before post four.
Read the posts. Try the workflows. Hire me to install them properly, or run them yourself. Both are good outcomes.
Connect on LinkedIn for notification when each post ships. No newsletter signup yet — that's a Phase 2 decision logged on the decisions log.