Writing
Notes on building with AI, and on getting it adopted.
Most of these start as something I posted on LinkedIn, and are argued out more fully here.
- Essays, not postsEach argued out past the thing that started it
- Adoption, agents, analytics, small modelsWhat the writing covers
- Always linked to the originalEvery piece points back to where it began
Showing all 10 essays
5 min readFrom asking to orchestrating: four stages of AI adoption
Ask, Automate, Delegate and Orchestrate describe where an organisation sits with AI and what each stage asks of its leaders, with a real case for each.
Read
4 min readTargeting is not dying, it is moving
Ad platforms may now choose who sees an ad better than you can, which leaves marketers with positioning and with checking the platform's own reported results.
Read
5 min read"The AI part was the easy part": a three-week quotation process
A manufacturer's quotes took two to three weeks and five in 150 became orders, and the technology was quick to build while changing how people worked was the real project.
Read
4 min readOpen weights remove one kind of lock-in, not all of it
You can keep an open-weight model you downloaded, but not the lab's roadmap, so due diligence on a model should include who funds the lab and whether you could run this version for two years.
Read
5 min readCheaper tokens, broken budgets: why AI agents need spending caps
Google now lets a hard spending cap pause its work agent, which tells us agents should be run as cost centres, with an owner, a limit and a named person who can press resume.
Read
4 min readYour default AI model may be one tier too big
A small model's reported jump on a computer-use test is a reason to re-test your own workloads on the smaller tier before you keep paying for the larger one.
Read
4 min readWhere enterprise AI stalls: between the pilot and daily use
Anthropic's $100 million academy suggests the hard step in enterprise AI is getting from a pilot that works to a system people use every day, and that depends on people more than tools.
Read
4 min readShoppers will let AI fill the cart, but not press buy
DoorDash's own figures and an Accenture survey point to the same design goal: an agent that does the work up to checkout and then makes the final review fast.
Read
4 min readAgree the measure before the budget: lessons from ChatGPT ads
A big launch proves that a channel exists, not that it works, so a conversational ad format needs a measure and a test agreed before any spend.
Read
5 min readChoosing the right model for the job: a layered approach to token costs
Using the most powerful AI model for everything is the fastest route to a large bill, so a layered approach matches each task to the cheapest model that does it well.
Read
Nothing here matches that. Try a broader word, or clear the filters.