Give four agents the whole problem and you get four stacks, four ideas of done, and nothing you can compare.
Give each one a single job, the same blocks and the same recipe, and the work merges!
Gordon Ramsay would recognise it: every plate that comes to the pass has to meet the standard. Period. I’ve been on about this for a while - for example: https://t.co/DIoGEBusR7 and https://t.co/sHGTvMs1T2
The video is a clip from my Boston AI Week talk, “Building an AI Software Factory with a Human in the Loop.”
#AIAgents #HumanInTheLoop #SoftwareFactory
Enterprise outages like the Microsoft disruption we’re seeing today are no longer just “email is down” events. They expose a much larger architectural reality: many enterprises are far more dependent on Microsoft than they realize, and that dependency often extends well beyond direct Microsoft contracts.
It’s not just Microsoft 365, Teams, Outlook, Azure, or Copilot. It’s the SaaS platforms, industry applications, identity systems, automation workflows, managed services, and partner ecosystems that quietly sit on top of Microsoft infrastructure. When Microsoft has a major issue, the blast radius can be much larger than the logo on the invoice suggests.
Reuters has a good summary of the outage here: https://t.co/mHOxPe2GjY
This is going to force a serious conversation inside enterprise IT. Not everyone is going to move off Microsoft tomorrow, and many probably should not. But the question will shift from “Is Microsoft reliable?” to “How much of our business is concentrated on a single dependency chain, and what is the cost when that chain breaks?”
I expect we’ll see more companies reassessing Microsoft exposure, especially in the platforms and vendors they rely on indirectly. Microsoft migration work will grow, but so will migration away from Microsoft where the concentration risk is too high. More importantly, repatriation is going to become a serious topic again. Not as a nostalgic return to the data center, but as a pragmatic discussion about control, resilience, latency, cost, and architectural optionality.
My consulting focus is moving in that direction: Microsoft migrations, migrations from Microsoft-dependent platforms, cloud exit planning where it makes sense, and repatriation strategies that are based on business risk rather than cloud religion.
The lesson is not “Microsoft bad.” The lesson is that enterprises need to understand their dependency graph before an outage draws it for them.
I did not write Tokenome AI... John O'Neil aka JON did.
In August he had something that worked and no clear idea what to do next, and what I contributed was.. telling him to ship it!
Last month he gave a talk about it, and Claude helpfully cut it to 37 minutes ... eight sections of two to five minutes each.
Here is the part I want to point out... In the evaluation section John tells you that two of our ten test questions did better without our product, and we could not reliably detect a change in answer quality, because it was lost in the noise.
He tells you that one evaluation is not a bulletproof examination.
We are a small company selling to engineers, and engineers find out. The only version of this that survives is the one where what we say in a talk ... works when you install it.
My talk is about the team server, and it exists because of something I watched happen. When everyone works through their own agent, reasoning stops being something you overhear. It sits in one person's chat window, in one person's account. The decision still gets made though, you just can't remember why.
So: one search across your own history and everyone's opted-in projects, with every result carrying the name of the person it came from. You pick which projects to share. Secrets are redacted on your laptop before anything leaves it. We do not host it, period. You run it.
It is in beta, and free now. Talk in the first comment.
As a solopreneur, you’re chasing a different goal every week: A deck. A website. A demo. A memo.
And then suddenly... success!!
Someone wants to place an order.
Now the demo has to get real. Customer accounts, so each customer only sees their own data. Payments without warning messages. Something that doesn’t fall over when customers two and three show up.
And it has to be ready by the date you just promised.
There’s no dev team to hand it to. Hiring takes months you don’t have. And every hour you spend wrestling the code is an hour you’re not spending with the customer who just said yes... that’s the moment https://t.co/DLPtKdNqG1 is for.
Take the order. We’ll handle the software.
https://t.co/szI1Ip0fUz
#smallbusiness #solopreneur
- Restaurants have no moat.
- Anyone can vibecook their own meal.
- You can build your own burger at home at a fraction of the cost it takes you to get it from a restaurant.
- If a restaurant builds a really good recipe, copycats will instantly reverse taste it and steal it.
Restaurants are dead.
JUST ANNOUNCED: RUSH “Elevate From The Norm” live residency at Sphere on May 21, 23, 26, 28, & 30, 2027!
*Register now* for Artist Presale at https://t.co/TUO3VRh1D2. Artist Presale begins Wed, Nov 11 at 10am PT.
VIP/Travel packages are available beginning Fri, Nov 6 at 12pm PT.
Tickets go on sale to the general public Fri, Nov 13 at 10am PT.
Claude Duet reads your connected apps in place, checking them every few hours. No index, no copy. Welcome to federated AI, Anthropic. The coffee is over there.
Two things to notice. Duet works with Claude only. And it runs in Anthropic’s cloud, for individual Pro, Max and Ultra plans; Team and Enterprise are not included yet.
For a business that cannot send its questions to someone else’s cloud, or chose its model on purpose, SWIRL is the answer. Live, permission-trimmed access to your own systems, ranked and cited, inside your environment, with any model. Claude included.