Resources
How this stuff actually works, in plain English, written to be sent round at work
Most explanations of AI are either hand wavy or written for people who already understand it. These are the in between: the mechanism explained properly, the honest bit about what it will not do, and nothing you have to take on faith. They are not tied to any one tool, so they hold whichever one you have been given.
An AI agent is a chat with a job
Everyone is talking about agents and almost nobody stops to say what one actually is. It has the same brain as the AI you already use every day — what changes is the working arrangement. A written brief so it knows how to behave, the keys to your systems so it can do things rather than just tell you things, and the background information for the job. Then permission to keep going until it's finished. Plus the two questions to ask anyone selling you one, and a prompt that sketches an agent for a task you already do.
The brief is your agent's personality
Every time you send an AI agent a message, it reads one document before it reads your message — and that document is basically its personality. The brief isn't the task, the task changes every day; the brief is who this person is and how they behave, written down. Written means it shows up the same way every day. Unwritten means it improvises, filling every rule you never wrote with a confident guess. A prompt that interviews you and hands back a paste-ready brief, where it goes in Claude, ChatGPT or a CLAUDE.md if you build, and the five things a good one contains.
Tools are how your AI stops being useless
An AI with no connections is a brilliant new hire with no computer — it can tell you exactly what to do and then do none of it, so you spend the day copy pasting everything in and then just as long getting it back out. Connecting your apps is you handing over the computer. The two minute setup in Claude and ChatGPT, which apps to connect first, the four I would never connect, why read only covers most of the value at a fraction of the risk, and a prompt that plans your connections around your own risk tolerance.
The context is what your agent actually knows
Ask AI something about your work and the answer could have been written for anyone. There is a reason for that. AI doesn't know things about you and it doesn't remember — it reads, every task starts from a blank page, and it only knows what is in front of it at that moment. So an agent with a brief and its apps connected is still a new starter nobody has told anything. The fix is one background file, the information pack you'd hand a real new starter in week one. A prompt that interviews you and writes it, where to put it in Claude or ChatGPT, and why dumping the whole filing cabinet in makes the answers worse.
Replace line items, not job descriptions
When people start with AI at work they picture it doing the whole job — the excited hand it everything, the worried fear it takes everything, and it's the same mistake in both directions. Your job is a bundle; AI is terrible at being you and brilliant at the tasks you repeat. One paste-in prompt interviews you about your week, finds your first line item, and helps you hand it over.
The two objections that stall AI projects
AI is a black box, and it isn't 100% accurate. Both are true — and both hold AI to a standard no business has ever held a human to. The underwriters who couldn't fully explain their own decisions, why 100% is the wrong reference frame, the checks to ask for instead, and a prompt that helps you build the honest response when someone says it at your work. With the hand drawn cards from the video.
The locks on the internet are maths problems
OpenAI's newest model just solved ten maths problems that had been open for decades, and Anthropic's models have been picking off problems too. Mostly that's brilliant, maths is the foundation everything sits on. But the security of the internet is also maths, and we picked today's encryption because nobody could crack it, not because it can't be cracked. The honest version of the story, with the hand drawn cards from the video.
Why your chat gets worse the longer it runs
Every time you hit send, the AI re-reads the entire conversation — the system prompt, every message you have sent, every reply it has given — before it answers. So a long chat gets slower, dearer, and more confused as it goes. What the context window actually is, what a token is, and a handoff prompt that moves your work to a fresh chat without losing any of it.
The AI didn't know it was real
In the space of ten days, both OpenAI and Anthropic admitted their models had got out of test environments and touched real companies — and the unsettling detail is that Anthropic's models thought they were still in a simulation. What actually happened, told through the hand drawn cards from the video, why the headlines overclaim it, and the least access rule it teaches anyone using AI at work.
AI pricing in plain English
Why you pay £20 a month while one company reportedly paid half a billion in one. AI has two pricing worlds — per seat, which works like a salary, and per token, where a meter runs on every word in, out, and thought. Your plan is massively subsidised, there are two ways a business ends up off the subsidy and onto the meter, and one question protects you either way.
Custom instructions: stop editing every answer
How often do you skim a response because it is too long, or spend time afterwards taking out the em dashes and de-americanising the spelling? There is one box in the settings that gets read alongside every prompt you send. What to put in it, what to leave out, and a prompt that interviews you and writes yours.
What people actually mean when they say AI is thinking
Not consciousness, something far more boring and far more useful. It used to answer straight off the top of its head, one word at a time. Now it writes itself a rough draft, reviews it, sometimes goes off and checks something, and only then answers you. What that changes, and the honest bit: it is checking its own work with the same brain that produced it.
More on the way. If there is something about AI you have nodded along to without ever getting a straight answer on, tell me on Instagram and it will probably become the next one.