If you’re expecting your AI agent to just “handle the data stuff” (scrape records, update deals, keep everything in sync) you’re wasting money and getting inaccurate results.
Use the AI for what it’s good at, and give the agent the tools it needs. Here’s why and how:🧵
1/6
Having spent this week building an agent that uses Luna for the majority of its work, Terra for self-QA, and Sol as an arbitrator for when the two disagree, this is fab news.
Not quite the rumoured “intelligence too close to meter”, but definitely getting there.
We are committed to pushing the model frontier across cost efficiency, capability, and speed.
Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.
Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.
@melvindvivas Now tested and (unsurprisingly) it doesn't sync Claude-to-Cursor. Once you start the conversation in Cursor, you'll need to keep it there.
@melvindvivas Gave it a go. And then noticed that it keeps the chat in sync moving forward.
I was still chatting to Claude and noticed the Cursor copy of the conversation updating in near realtime.
Haven't tried to see if the conversation updates in Cursor update Claude.
Very much agree with the need for more technical education.
But let's not make ourselves vulnerable to the next tide of the coming wave. AI leads to dramatic improvements in robotics, which will impact this sector too.
This policy works provided it's part of a rounded education, and a re-vitalised tertiary education sector. And an economy designed to retain the companies we build in the UK, rather than making selling to (or listing in) America the most attractive option.
Claude Code → Cursor import is live.
After you import, verify:
1. Re-open one real project chat
2. Ask a follow-up that needs prior context
3. Trigger one skill you rely on
If step 2 fails, the move wasn't complete.
I was planning to get my whole team on Claude Code in VSC but noticed most were still using @cursor_ai - which I thought was hugely outdated (I hadn’t used it for around 18 months).
Resisted the temptation to demand change until I was confident I knew what I was talking about. So, I reinstalled Cursor for myself and used it on a few small projects.
It’s so bloody good.
Awesome developer experience, superb harness, great cloud agents, incredibly cost effective, and obviously moving in a great direction.
So, what did I go?
Upgraded everyone’s Cursor seats to the Premium tier, switched on BugBot, and encouraged them to use it even more.
Always check why your team is doing what they’re doing before you act.
@tombo19722@KanishkaNarayan Not sure how you get to that from the points I’ve made.
But I accept that those are the things that we need to avoid while taking the steps to make people better off.
Opus 5 launched and the benchmarks are ludicrous.
The jump in @claudeai’s knowledge work performance is remarkable. And the novel problem solving capabilities are astonishing.
Now to find out if it performs at that level in real world use…
We have today appointed my brilliant colleague, Lord Vallance, to chair the PM’s AI Taskforce.
Patrick and I worked closely together before. We will now do so again, with the urgency and importance of the Vaccines Taskforce, applied to the central question of artificial intelligence.
https://t.co/rCDWoBNkrO
Spot on (as Ryan usually is).
In Basecamp we intentionally don't use the word "priority" anywhere in the product. Yet we have tasks, assignments, cards, to-dos, all the things that would normally be prioritized.
Marking something as high priority, or ranking it 1-5 or something like that, is a painful trap sprung by software on unassuming teams and project managers.
Same as % of a list being done equaling some sort of progress. It's not progress, it's just a list getting done. 9 out of 10 things done on a list doesn't make something 90% done. An indicator of "# of things done" is fact, and that's fine, but representing it as progress is not.
Beware of pseudo-labels that are easy to apply. They mask deeper meaning you then ignore.
@ProfBernardPaz@KanishkaNarayan@andyburnham Or we could assume best intent and strive to deliver a great result for our local and national communities. I definitely prefer that option.