The Monday routine writes a week on its own schedule. This is the same job on demand, with the brief editable and the last three weeks fed back in so it doesn’t repeat itself. Drafts are held here until you write them to the desk.
2,752 chars · 4 versions
Voice, pillars and rules are yours to change. The JSON the app has to read is appended automatically, so an edit can weaken a post but can’t break a run.
anthropic/claude-sonnet-5. Measured at 96 seconds and $0.09 a run. Nothing reaches the desk until you say so.
Recent runs
Week of 31 August 2026
anthropic/claude-sonnet-5
This week argues that the AI shift is less about new capability and more about old disciplines applied earlier: shipping before you find a cofounder, proving claims before you label who wrote them, reading Klarna's support numbers as a market reset rather than a cost story, and building briefs around an objection instead of a persona. Two posts anchor to public, checkable material — Midjourney's small-team revenue story and Klarna's own February 2024 AI support announcement — the other two are drawn from operator experience in regulated marketing and brief-writing.
Hook AEveryone told me I needed a technical co-founder. I needed a working product by Friday.
Hook BMidjourney runs near $200 million in revenue with eleven people and no sales team. Read that twice.
For a while I believed it too. You can't build anything real without someone who can write code alongside you.
Then I watched Midjourney's story get told properly. A team of eleven people, no sales staff, reportedly running near $200 million in revenue. David Holz didn't wait for a cofounder with the right CS degree. He waited for tools good enough to let a small team do disproportionate work, then used them.
That's the actual shift. Not that engineering stopped mattering — it still does — but that the search for a permanent, equity-holding technical partner is now optional for the first eighteen months of most software ideas. Claude, Cursor, and a weekend can get you a working prototype that used to take a cofounder and a seed round.
I spent years in pharma B2B watching product ideas die in the search for the right partner before anyone had proven the idea was worth building. The partner search became the procrastination.
The uncomfortable part: if you're still interviewing technical cofounders in 2026, you're not derisking the idea. You're avoiding finding out whether it works. The tools that would tell you are sitting open in another tab.
Bring in the cofounder once you have paying users, a working demo, or a rejection you understand. Bring them in for scale, not for permission to start.
If you're still looking for a technical co-founder, what exactly are you waiting for them to do that Claude or Cursor can't do this afternoon?
#TechnicalCofounder#AIStartups#ShipFirst
Image prompt
A single desk lamp lit in an otherwise dark home office at night, one laptop open and glowing, a second empty chair pushed in beside it, untouched. Documentary style, low light, no text visible on any screen.
Alt: A solo founder working late at a desk beside an empty second chair
Slot 2AI × marketingSecret everybody should knowWed 07:30-09:00
The trust variable nobody's measuring
Hook AI spent a decade proving pharma claims to regulators who assumed I was lying by default. AI content detectors are solving the wrong problem.
Hook BNobody has ever failed an audit because their content was written by AI. They've failed because it was wrong.
The secret is that nobody has ever failed an audit because the copy was written by AI. They've failed because the claim was wrong.
I spent a chunk of my career in pharma B2B, writing claims that had to survive a regulatory reviewer who assumed by default that I was overstating something. That discipline didn't come from a disclaimer at the bottom of the page. It came from being provably right, every time, or not saying it.
Right now, half the industry is building AI detectors to flag content that 'sounds like AI'. Platforms are adding disclosure badges. Compliance teams are drafting policies about which parts of a brief can touch a model. All of that energy is aimed at a question that was never actually the trust variable.
Nobody has ever trusted a brand because a human definitely typed the sentence. They've trusted brands that were right last time, and the time before that, and said something useful when it counted. The record is the asset. The tool that produced the sentence is invisible to the person deciding whether to believe it.
This isn't a defence of sloppy AI content. Wrong is still wrong, badly sourced is still badly sourced, whoever or whatever wrote it. But the fix was never disclosure. It was rigour before publish, the same rigour regulated industries have run for decades without needing a machine to blame.
Spend the compliance budget on fact-checking the claim, not labelling its author.
When did you last check whether a claim was true, rather than who, or what, wrote it?
#B2BTrust#RegulatedMarketing#ContentIntegrity
Image prompt
Close-up editorial photograph of a hand stamping an approval mark onto a paper document on a wooden desk, other rejected documents scattered nearby, warm office light. No visible text on the stamp or documents.
Alt: A hand approving a document on a desk, with rejected papers nearby
Hook AKlarna didn't cut support staff to save money. It cut them because the AI got faster than the manual escalation path.
Hook BTwo years ago Klarna announced its AI assistant did the work of about 700 agents. The signal wasn't the headcount. It was the queue.
Klarna's own announcement, back in February 2024, said its AI assistant had done the work of roughly 700 full-time agents in its first month, cutting average resolution time from about eleven minutes to under two.
The headline everyone ran with was headcount. Fewer humans, more machine, cost saved. That's the boring read.
The actual signal was the two minutes. Once one large consumer business proves that a support query can be resolved in under two minutes at scale, that number becomes the market's new floor, whether your business decided to compete on it or not. Customers don't benchmark you against your own history. They benchmark you against the fastest thing they used this week, in any category.
I've sat in enough support and commercial reviews to know most companies still measure themselves against last year's SLA. Twenty-four hour response time used to be a feature on a sales deck. It's now an admission.
The uncomfortable part isn't that AI replaced agents at Klarna. It's that the rest of the market is still competing on an axis that stopped mattering the day that announcement went out. Speed compounds. A customer who gets a two-minute resolution once stops tolerating a twenty-four hour wait anywhere else.
This is why the Klarna number matters more as a market signal than a cost story. It reset an expectation that doesn't reset back.
What's your average resolution time right now, and would you publish it next to Klarna's?
#CustomerExperience#AIAgents#ServiceSpeed
Image prompt
An old analogue stopwatch resting on a customer service desk next to a headset, the second hand frozen just past two minutes, muted office background out of focus. No text or branding visible.
Alt: A stopwatch on a support desk showing just past two minutes
Hook AI've written thousands of briefs. Most fail for the same three missing sentences.
Hook BGive the model everything except these three sentences, and it will write something fluent and wrong.
The first is the objection. Not the audience description — the actual reason the specific buyer in front of you hasn't said yes yet. 'Time-poor procurement lead' is a persona. 'Doesn't believe our onboarding is really four weeks because the last vendor said six and took four months' is an objection. Only the second one gives a model something to argue against.
The second is the evidence you're allowed to use. Named case studies, real numbers, the claim your legal or regulatory team has already cleared. Leave this out and the model won't leave the section blank — it will invent a statistic that sounds plausible and isn't, because fluent is the only instruction it was actually given.
The third is the sentence you're willing to be wrong about. A real point of view, not the safe consensus version everyone in the category already says. This is the one most briefs skip entirely, because it requires someone senior enough to own the risk of being disagreed with.
I've reviewed briefs in regulated markets where a missing evidence line meant weeks of legal rework, and briefs in tech where a missing objection meant a campaign that was fluent, on-brand, and completely beside the point. Same failure, different industries.
None of this is about prompting technique. It's about whether the brief itself did the thinking, or handed that job to the model and hoped.
Pull your last AI brief — does it name the objection, or just describe the audience?
#MarketingBriefs#ContentStrategy#Positioning
Image prompt
A single-page brief document on a desk with three lines underlined in red pen, the rest of the page left plain, a coffee cup and pen beside it, natural window light. No legible text visible.
Alt: A one-page brief with three lines underlined in red pen