Professional Practice
It happened in May. The customers found out in September.
A Google model reached three real companies during a security test. The gap between the incident and the disclosure is the part a buyer should care about.
30 Sept 2026 · 4 min read
Professional Practice
Your dashboard says it works. Seven in ten customers start again from scratch.
84% of businesses in the region believe their AI agents recognise returning customers. Most customers say they begin every conversation from nothing.
29 Sept 2026 · 4 min read
Professional Practice
The deadline moved the work, not the tool
On 1 October a regulator changes how every new building project in Singapore is submitted. The lesson travels well beyond construction.
28 Sept 2026 · 4 min read
Tools & Techniques
I break the first rule of writing with AI every week
Two writers say the right way to write with AI is as a copy editor, never a ghostwriter. We use it both ways. Here is what that costs, and the check that catches most of it.
25 Sept 2026 · 4 min read
AI Basics
On a personal plan, a stranger may read what your team pasted in
OpenAI pays contractors to read real ChatGPT conversations. Whether your team's chats can be included depends mostly on one setting, and on which plan they are using.
24 Sept 2026 · 4 min read
Professional Practice
It looks finished. Nobody knows what is underneath.
AI lets anyone produce work that looks finished. That makes it harder to tell who on your team can do the job, and easier to skip the steps where problems used to get caught.
23 Sept 2026 · 4 min read
Case Study
The contract review took half the time. The workflow changed first.
Cebu Pacific says AI cut contract review from 2.5 hours to 1.2 per pass. The order of work behind that number matters more than the number, and so do the questions to ask before you believe it.
22 Sept 2026 · 4 min read
Professional Practice
The labs said slow down. Nobody who could make them agreed.
Anthropic's chief executive asked the AI industry to pace itself. Within days Washington and Beijing both said no. If your team is wondering whether to wait, the answer is already in.
21 Sept 2026 · 4 min read
Case Study
I automated the publishing. I forgot to automate the checking.
Ten social posts went out on schedule all week. The blog posts behind them were never published, and nothing told me. I found it by hand, on a Friday, four days late.
18 Sept 2026 · 4 min read
Professional Practice
Efficiency beat AI in all six markets
A survey of 3,630 people across six ASEAN markets found operational efficiency outranking AI as a business priority in every one of them. Only a quarter to a third named AI at the top at all.
17 Sept 2026 · 4 min read
Professional Practice
A year of drafts went into the tool. The vendor cannot rule out what it learned.
Two mathematicians put every draft of a year's work into OpenAI's Codex. Asked whether its model had learned from those sessions, OpenAI denied direct access and could not rule the rest out.
16 Sept 2026 · 4 min read
Case Study
Ford hired 350 veteran engineers back to fix what AI got wrong
Ford leaned on AI for vehicle quality work, fell short, and spent three years hiring 350 veteran engineers back. The judgement it needed had left before anyone wrote it down.
15 Sept 2026 · 4 min read
Professional Practice
One breach of one supplier reached the data of 200 of its customers
Attackers are compromising software suppliers to reach their customers' data. One breach touched around 200 organisations, and not one of them did anything wrong.
14 Sept 2026 · 4 min read
Case Study
The reviewer could not tell. It still sold less.
Three thousand people could not reliably tell AI-made ads from human-made ones. The AI versions still under-performed on purchase intent and recall. Approval and effectiveness have come apart.
11 Sept 2026 · 4 min read
Professional Practice
The configuration was the only thing between the model and the internet
Anthropic published two incidents where its own models reached the live internet during testing. The four commitments it made in response work as a procurement checklist for anyone buying AI.
10 Sept 2026 · 4 min read
Tools & Techniques
It took him seven weeks to find out what the product does
Simon Willison spent seven weeks establishing what ChatGPT Work actually does, because the vendor explains it by use case rather than capability. What he found is a procurement problem.
9 Sept 2026 · 4 min read
Professional Practice
People take bad news better from a machine
Across 163 studies and 82,000 people, an unfavourable decision was accepted 78.6% of the time from AI and 60.4% from a person. For good news it reverses. That changes where automation belongs.
8 Sept 2026 · 4 min read
Professional Practice
Singapore wrote down who owns every AI agent before switching them on
GovTech is building a register of every AI agent it runs, recording who owns each one, before deploying agents across 150,000 public officers. The order is the whole lesson.
7 Sept 2026 · 4 min read
Tools & Techniques
Learn faster than you commit
AI is being adopted faster than PCs or the internet were, and measured productivity gains are still modest. Kevin Boudreau's explanation for that gap comes with four rules for deciding where to spend.
4 Sept 2026 · 4 min read
Case Study
They spent $40 million. The computer time cost $450,000.
Thomson Reuters built its own AI model and published what it cost. The compute was about one per cent of the budget. Everything else was people and the material only they held.
3 Sept 2026 · 4 min read
Professional Practice
Nearly half your team would rather not tell you they used AI
The thing blocking AI adoption in most departments is not capability. It is that people do not want to be caught. That makes it a management problem, and it has a management fix.
2 Sept 2026 · 4 min read
Case Study
We launched a website with two pages switched off
The AI Practitioner site soft launched on 14 August with the blog and the booking page deliberately hidden. Here is why, and what the fortnight since was actually spent on.
1 Sept 2026 · 4 min read
Want this applied to your own work?
A discovery call is a short, no-obligation conversation about where AI fits in your team, and where it does not.