- The AI Optimist | AI news and strategy for leaders
- Posts
- Human is the word
Human is the word
The more AI can do, the more human work matters.
Friends,
your weekly AI briefing is here - designed to help you respond to AI, not react to the noise. No curveballs. No chaos. Just clarity.
The wand got cheaper again: new models cost less, agents learned to pay, and AI that once needed a server room now runs on one desktop computer. So we ask the bigger question: what is work for?
Cambridge Dictionary has chosen human as its Word of the Year 2026, reflecting a renewed appreciation for human creativity, connection and authenticity in an AI-driven world.
A quick note from Hugo
On 7 January we're bringing a small group of curious people together at a secret location in London for a BrainRave. Expect human magic and AI; the rest you discover on the night. Reply BRAINRAVE if you want to start 2027 with your brain fizzing.
📰 This was the week that was...
This was the week the AI wand got cheaper.
The cost of AI doing a piece of work keeps falling. The more interesting question is what we do with the human time and attention that creates.
Agents are starting to spend money too. On 13 November OpenAI launched ChatGPT Wallet: its agents fill a basket and check out for you, with your approval for each purchase. Two days earlier United Airlines said it will block unauthorised third-party AI agents from booking flights, in favour of approved, secure connections. The rules of the road for agents are being written now.
The week's biggest story was of course about a person. On 12 November Jensen Huang resigned as Nvidia's chief executive after 33 years. The company behind much of the AI boom's chips moved on one human story. Meanwhile Anthropic, the maker of Claude, began trading on the stock market on Thursday. AI's biggest players changed shape in one week, and even the AI economy turns on people.
Cheaper AI wands, new rules and very human headlines: what will we do with the human time and attention that cheaper AI creates?
Let's get into it.
🔥 Urgent Priorities
This week's priority is cyber: new cyber attacks are happening every week. Is your organisation focused enough on it?
Three break-ins carried out by AI agents in eight days: each came through an ordinary gap: a test login left in a live system, a system left open to the internet, or no cap on how often it could be used. No new style of attack was needed.
On 13 November the FBI's advisory on ARTEX, a free security-testing tool now misused by attackers, says to check the least-watched systems: supplier portals, older services and third-party connections.
✅ Worth doing this week: book half an hour with your IT lead or security partner and ask: which system open to the outside world gets the least attention, when was it last checked, and are any test logins sitting in live systems?
Cyber isn't new. AI amplifies the risk, so as leaders we need to amplify our focus.
🎯 Strategic Insight
The Cambridge Dictionary's word of the year is human. That feels timely. As AI makes producing things cheaper, we need to become clearer about the work only people can own: judgement, responsibility, meaning and connection.
The No AI Slop manifesto gives us a useful test. It is a practitioner manifesto, not a research study, but its five values are hard to improve on:
Outcomes over output. Did this help anyone, or did it simply create more material?
Understanding over speed. Can we explain what is happening?
Our own thinking over a model's first draft. What did a person decide before the model started writing?
Better work over easier work. Did AI improve the thing, or just make it quicker to produce?
Improving the system over patching the output. If the same mistake appears twice, what should change upstream?
The tension is simple. When leaders ask, “Which roles can we automate?”, people hear, “What is my work worth?”
A better question is: what would this person be free to do if AI took the routine?
That is the human opportunity. As the cost of doing falls, we can invest more deliberately in the things that make work worth doing: wisdom, energy, identity, standing, connection and presence.
Pick one piece of AI-assisted work from this week and run the five-question check. If the answer to any question is uncomfortable, do not immediately ask for a better prompt. Ask whether the work, the process or the system needs to change.
🤓 Geek Out
A small hospital model beats the AI giants at running a hospital
NYU Langone Health built Lang1, trained on the hospital's own health records plus internet text. Once trained for each job, a compact version beat general-purpose AI up to 70 times its size at predicting 30-day readmission, mortality, length of stay and insurance claim denials, and it worked at a different health system too.
Why it matters: a small model that knows your world can outperform a giant that knows everything else. Your organisation's own records may be its most valuable AI asset.
👉 Action: list three predictions your team makes by gut feel each week, and the records you already hold that could inform them.
AI that chooses instead of writes
On 11 November Google launched Gemini Decision 1, one of a new wave of “decision models”. Instead of writing text, it scores the options you give it. One request can ask up to 1,000 questions about one document, image or call, with answers in about a fifth of a second, for a fraction of a penny a request. Perplexity, Liquid AI and HeyGen shipped similar models the same week.
Why it matters: much everyday work is sorting: which enquiry is urgent, which lead is warm, which invoice needs a second look. Decision models make that triage fast and cheap, with a person still making the final call.
👉 Action: write down one daily sorting job, its options and your rule of thumb. That page is the brief for a decision model.
Big-league AI on a single desktop computer
Edge0 has released a free AI for writing software and running agents that works on one well-specified desktop computer, with no cloud. AI this capable usually needs a rack of servers in a data centre; a smaller version runs on Apple Macs.
Why it matters: frugal AI keeps getting stronger, and AI that runs in your office keeps your data in the building.
👉 Action: ask your IT lead which AI uses involve data you'd rather keep in-house. Those are your first candidates for AI that runs in-house.
🎨 Weekend Playground
This weekend, design your dream room with ArchLang, a free, open-source language for writing floor plans as plain text. Released by Chan Meng in June, it has passed 18,000 downloads, and AI agents called it 16,800 times in the last 30 days.
Why this matters: AI works brilliantly on any world you can write down in a language it reads. Floor plans this weekend, your processes next.
Mission:
Take a tape measure round one room: walls, doors, windows and where the light comes in.
Describe it out loud to Claude via ArchLang's connector or ChatGPT via its app and ask for an ArchLang file. No AI? Type it straight into the free playground.
Paste it into the playground and watch your room appear as a precise drawing.
Now redraw it as the room you'd love to work in. Bigger window? Room for a plant? A corner for the dog?
Close the laptop and change one real thing in the room.
If The AI Optimist helps you make sense of what's happening, forward it to someone who's figuring out what AI means for their organisation.
Got a wish? Reply WISH and tell me the piece of work you wish was different. At AI Night School, we help leaders turn wishes into working systems.
Stay strategic, stay generous.
Hugo