Work past the first answer
Your Essential Report from the Frontlines of Applied AI
This week: Let's explore how to work with AI to increase the quality and uniqueness of your answers.
If you only use AI as a one-shot ‘answers engine’, it can indeed give you a decent looking draft in the first pass.
But to get real quality, never stop with the first pass — read it, critique it, and ask for the generation again. This process is what improves your answers.
In this issue: a useful prompting technique for the critique process, and a case study on this process carried out at scale in a Singapore hospital.
And of course the latest news this week in applied AI spanning education, law, healthcare and corporate IT.
AI gives you the ‘median’ answer. Everything that makes it unique to you and your needs, you will have to add yourself.
Most people never get that far. They ask, they read, they use it — the transaction model, as AI+ Academic Research names it. Quick, but often wrong, because the answer was written for a generic reader, not for you.
Brief it like an assistant instead. You know things it can't: the budget you're working to, what your reader has already been told, why the last version came back rejected. None of that is on the internet.
So put it in. Read the draft, name the thing that's wrong, add the piece it was missing, ask again.
By the third pass the answer holds things only you could have supplied. That is your unique personal value, crystallised in the AI's work. Truly AI+You.
- Education
Houston ISD and 54 more US school systems gave their teachers ChatGPT for lesson planning
OpenAI extended ChatGPT for Teachers to 55 school systems on 26 August, including Baltimore City, Northside ISD and Virginia Beach. Teachers use it for lesson planning, parent communication and administrative work.
EdTech Innovation Hub, 26 August 2026
- Legal
Greenberg Traurig put an agentic AI assistant in front of all 3,200 of its lawyers
CoCounsel Legal went firmwide on 26 August for research and drafting, including a Brief Builder that assembles a first draft.
Thomson Reuters press release, 26 August 2026
- Healthcare
An NHS hospital used AI to move urgent chest X-rays to the front of the reading queue
Queen Victoria Hospital NHS Foundation Trust went live with Harrison.ai triage on 26 August. Scans with significant findings are pushed up the list instead of waiting their turn; radiologists still read and report every one.
ResultSense, 26 August 2026
- Corporate IT
Cisco gave all 90,000 of its employees a personal AI agent that runs their workflows
MyAgent reached the whole workforce on 27 August. It works across Outlook, Webex, Jira and SharePoint, carrying out multi-step tasks rather than answering questions about them.
PYMNTS, 27 August 2026
Ask an AI to improve a draft and it hands back a different draft: your good opening gone, a heading you liked reworded, and the one paragraph you wanted fixed no better than before. Name the changes instead, one at a time, and tell it to leave the rest alone. The brackets are yours to fill in. If you aren't writing marketing copy, swap “copy” for whatever you are writing and drop the voice line.
You are a senior copy editor. Please review the following copy and improve it according to these specific criteria: 1. [List your specific revision brief — e.g. "Tighten the opening — it's currently 3 sentences and should be 1"; "Change the CTA from passive to active voice"; "The second paragraph is feature-led — rewrite it as benefit-led"] Voice calibration: [paste Voice Calibration Brief or key markers] Do not rewrite sections that aren't specifically noted for revision. Preserve the structure and approach of the sections that are working. [Paste copy to revise]
Prompt 3.18 from AI+ Copywriters, word for word.
Read every piece of patient feedback with AI, find what keeps going wrong, then change the process that causes it.
Alexandra Hospital, part of Singapore's National University Health System, collected 126,134 pieces of patient feedback through its portal in 2023. Around 92,578 carried a written comment — more than 1.5 million words. Nobody reads that by hand.
Sentiment analysis, topic modelling and emotion detection found the themes running through it. The hospital then fixed five of them: a queue notification system, communication training for staff, redesigned billing statements, sound-dampening and bedside changes, and separate service protocols for different groups of patients. Waiting-time satisfaction rose 18%, communication ratings rose 15%, billing complaints fell 23%. It is written up in JMIR Formative Research. There was no control group, so those are before-and-after figures.
The AI never answered a single complaint. It read the whole pile, the humans picked the five things worth fixing, and the complaints fell because the causes went away.
