Notes on design, technology and business
A running view on how customer needs, business reality, technology and organisational context connect, and what that means for how problems should be approached.
The terms used across this section are indexed at the vocabulary.
The AI opportunity audit, in practice
A worked example of the AI Opportunity Audit template, run across a real, multi-department retail business this week.
02Context Debt
The parallel to technical debt that AI is making impossible to ignore: undocumented decisions, tribal knowledge, and exceptions nobody wrote down.
03Context Design
Context design looks beyond the visible brief to the customers, technology, processes and business realities actually surrounding it.
04Context engineering vs. context design
Context engineering is getting the right material into a model in the right shape. Context design is deciding what that material should actually be. Almost everyone selling one is quietly assuming the other is already solved.
05Files and folders vs. agents
The infrastructure debate over filesystems and vector databases is the same argument as context design, one layer down the stack.
06Five patterns I keep seeing in large AI rollouts
Composite patterns observed across large-organisation AI rollouts, repeated often enough to be worth naming.
07How context gets designed
A worked example of curating context, using a single broken support ticket to show what the process actually involves.
08How this site is built
The site itself is a small case study in context design: plain files, no database, and nothing shown to you that isn't the real thing underneath.
09Is my business ready for AI?
Readiness usually gets measured in tools, budget and skills. The better test is whether your business has written down what the AI would need to know.
10The AI tool directory boom is a symptom, not a fix
Prompt libraries, tool directories and glossaries all solve the awareness problem. None of them solve the context problem, and that gap is where the disappointment lives.
11The job that doesn't have a name yet
A role is being built right now under other job titles, because the need showed up before the title did.
12The model can't hallucinate. Your inputs still can.
A new class of AI model guarantees every answer is well-formed. That solves one problem and makes another harder to see.
13Why AI-native companies need more people, not fewer
A structural argument, using three companies in one sector, for why cutting headcount is the wrong way to compete on AI.
14Why does AI make everything sound generic?
Bland, could-be-anyone output isn't the AI being lazy. It's the AI being accurate about the only thing it knows: what most businesses are like.
15Why prompt engineering was the wrong abstraction
Two years of advice about wording prompts better missed the actual lever: what material the model has to reason from.
16Why two teams get different results from the same AI
Same model, same subscription, wildly different output: the variable isn't skill at prompting, it's what each team feeds it.
17Why your AI rollout doesn't match the pilot
The pilot always looks better than the rollout. That's not a sign the tool got worse, it's the first honest look at how much curation the pilot was quietly doing the whole time.
18Why your team keeps re-explaining the same context to AI
Before anyone asks the AI for anything, they type the same paragraph of background. That isn't a tool problem. It's a list of what your business never wrote down.
19Your staff are already using ChatGPT. Now what?
In most small businesses, AI arrived without a rollout: people just started using it. The risk isn't that they use it. It's that everyone's using it differently.