Working notes from building Groundworks AI — on why marketing stalls for lean teams, and what changes when the system brings you the work instead of waiting for you to start it.
Half the web’s new articles and four in five long LinkedIn posts now test as AI, and readers mark down anything that feels machine-made. An opinion on the mistake behind it: tools built to replace you. And the design principle we took from it: the tool has to work to get to know you.
Read the post →An evidence review. Founders and small businesses say marketing matters and measurably don’t do it — 70% give it under five hours a week, and more than half of early-stage B2B start-ups do none. Why the gap has four causes, not one, why LinkedIn is the channel a company of three has left, and what the counter-evidence says. Seventeen sources, each read at source and graded.
Read the review →Every founder says the same thing about why their marketing went quiet: not enough hours. The tell says otherwise — approving a draft someone else wrote takes ninety seconds. The constraint was never time, and naming the real one changes what fixes it.
Read the post →A clean report has a short shelf life. Here is how we combine AI-assisted review, human judgement, regression tests and live production checks — and why the process has to be able to demand immediate action.
Read the post →Most advice starts from a premise nobody checks: that the goal is an audience. For a founder it isn’t. LinkedIn’s real job is to convince the people who already have a reason to look you up — and that reframe changes every decision downstream.
Read the post →Connection games, pods, posting-time folklore, hook formulas, format chasing. Here is what each one actually optimises for, why it isn’t your goal — and the four unglamorous things that are genuinely worth the time.
Read the post →Three decisions govern a whole quarter of output: the theme, the audience, and the lead framing. They get made per post, in three minutes, by someone whose attention is elsewhere — because skipping them is free on the day.
Read the post →Read your last twenty posts in one sitting. Individually they’re fine; together they’re one message wearing twenty outfits. That isn’t a discipline problem — it’s the predictable arithmetic of how they were produced.
Read the post →You don’t talk one way, and you can’t hear yourself — which is why the most common writing advice given to founders is useless. Voice, taken apart into four things you can actually work with.
Read the post →Readers make the judgement within a line or two, and once it has arrived the post is over. Eleven recognisable patterns that trigger it, what to do about each — and the honest limit of fixing them.
Read the post →Every founder does four good weeks. This is about week six — what a real week actually contains, why the pipeline breaks exactly when the company most needs to look alive, and how drift differs from a deliberate skip.
Read the post →Every item on the list is reasonable and the total is not. The whole series added up — five kinds of work on three different clocks, for a company where marketing is the eleventh priority.
Read the post →Every conversation about AI arrives at the same question: whose job does this take? It carries an assumption worth making visible — that someone is sitting in the seat. At most early-stage companies, nobody is.
Read the post →There's a comfortable story that AI hasn't fixed marketing for small teams because the models aren't smart enough yet. That story is wrong about where the problem lives. The bottleneck was never output quality — it's the blank box you have to face before any output exists at all.
Read the post →If you've generated a few drafts, felt a flicker of momentum, and then quietly stopped three weeks later — you're not undisciplined and the tool isn't broken. You're running into a structural problem hidden inside something that looks like a feature: the prompt box.
Read the post →