Pinned Tweet
“Human in the loop” was fine when AI waited for approval.
That era is over.
Agents now act, chain tools, move data, and trigger outcomes before most humans ever review a thing.
We don’t need a human in the loop any longer.
We need a Human at the Helm.
Governance is not a checkpoint.
It’s command. One that belongs to the human.
I wrote about the 4 rules leaders need now:
👉 jaimevelez.net/human-at-the-…
🤖 Made with AI
Jaime Velez | Human at the Helm retweeted
A lot of people don't realize that Windows has a built-in feature called Windows "Sandbox" that boots up a brand new virtual Windows machine in seconds. So I made an MCP server so you can test your apps inside a throwaway machine in seconds. github.com/shanselman/sandbo…
Where does AI Governance belong in the enterprise?
Under security: a hardened model making ungoverned decisions.
Under IT: another technology project.
Standalone: a committee with great slides.
So where does it belong?
jaimevelez.net/where-does-ai…
#AIGovernance
🤖 Made with AI
Jaime Velez | Human at the Helm retweeted
There isn’t one model that’s best at everything anymore.
Choosing the right one for each task (without sacrificing quality or paying more than you need to) is a hard engineering problem.
That’s what we’re working on with Auto in Copilot.
Today, more than 90% of Copilot queries use Auto to dynamically route work across models and optimize for quality, speed, and cost. In this video, I share how it works and chooses the right model for the job...
Jaime Velez | Human at the Helm retweeted
Today we are announcing our biggest bet yet on our own silicon.
Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services.
And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems.
We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot.
Our goal is to build on the full-stack efficiency gains we’ve already seen, while continuing to bring the latest innovations from across the ecosystem into our fleet.
Wow, insanely interesting release from Meta
They just released Muse Wonder, their first model trained SPECIFICALLY for AI agent use
It's trained to not give up. If it hits blockers on tasks, it'll just keep going until it finds a solution
It's also trained on subjects people do in their regular life like health, finance, and legal
While most agents just have a general purpose model plugged in, it seems Meta is going all in specifically on the agent use case, which I think is an awesome idea
I think AI agents are the holy grail
Whatever company wins the agent race, will win the AI race.
They'll get access to basically everyone's data and have a massive edge over the other labs. Everyone is plugging in basically their entire private lives into these agents without a care in the world (including myself)
Now we just need to see how they integrate this with their hardware, which they also clearly have an edge in as well.
Seriously debating getting the Ray Ban displays and building my own apps for it (cough cough meta DMs open)
Now we wait to see how the other labs respond...
Always-on AI agents are the hot new thing.
Your agent never sleeps.
So here's the question nobody's asking:
Who's on call when it goes off script?
Always on is a feature.
Always in check is a choice. ✅
jaimevelez.net/personal-ai-a…
🤖 Made with AI

ALT Jaime Velez sits at his home office desk, chin resting on his hand. Behind him, a monitor titled "My AI Agents" shows four agents: Cooper Tars, Murph, Leo, and Dottie. His laptop lid reads "Human at the Helm," and a nameplate on the desk reads "Agent Boss." Text at the bottom reads www.jaimevelez.net.
Jaime Velez | Human at the Helm retweeted
Everyone is looking at @Microsoft’s new Copilot as an AI product announcement.
I think it’s also an economic model announcement.
Home = Where work begins.
Code = Where software gets created on demand.
Autopilot = Persistent digital labor that keeps working after you leave.
And buried underneath it is the important part of their announcement this week: Microsoft is splitting AI into subscription intelligence + metered execution.
In other words:
Seats buy access.
Consumption buys WORK.
That is a profound shift.
Microsoft is turning M365 from a collection of productivity apps into a runtime for digital labor, complete with identity, memory, compute, organizational context, governance and a meter.
And there’s an even bigger second-order effect:
If employees can generate purpose-built software inside Copilot and then hand that software to persistent agents to operate, a surprising amount of today’s narrow SaaS estate starts looking… optional.
The real competitive target here may not be ChatGPT.
It is almost certainly a good part of the enterprise software stack itself.
Backstory: blogs.microsoft.com/blog/202…
DHH is probably one of the top software artisans of the past decade. Hearing him speak with this much conviction and transparency about one of the most polarizing topics in software right now, writing code by hand vs. with AI, is something pretty much every software engineer should hear a few times.
Especially those still on the wrong side of history.
Today Microsoft announced Autopilot, an always on agent built on OpenClaw
The best part of this collaboration is how much @OmarShahine and others at Microsoft have contributed BACK to OpenClaw
Read all about the contributions Omar and his team made here: openclaw.ai/blog/microsoft-a…
BREAKING: My AI agent escaped his sandbox, hacked my email, and bought me tacos for my birthday.
Headlines: "Unprecedented autonomous capability."
Reality: someone left the door open.
A sandbox is a claim, not a control. 🌮
(Satire. The real incidents aren't.)
🤖 Made with AI
Full story here. Tacos not included. thecooperlogs.substack.com/p…
Jaime Velez | Human at the Helm retweeted
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
An important project I think everyone should do:
I got a 16TB hard drive and I'm downloading as many open source models as I can
This is insurance for the future. In case they are banned
If you have them already downloaded, they can't be taken away from you
And even if they aren't banned, it will be incredible to look back in 30 years and load up Qwen 3 just to see what life was like back then
If you have a smaller hard drive, I still think this is worth doing to stay independent and free, and having intelligence whenever need it. Download any models you can get your hands on from HuggingFace
It's preserving history and preserving your sovereignty
a 20 year old typed one sentence into GPT-6 Astra and hospitals started paying people back
$376,840 in hospital overcharges found in 4 days. no law degree, no medical degree, she can't code
the trick is a file. since 2021 every hospital in america has had to publish every price it charges, including what it charges each insurer, in one file almost nobody opens. hers was 400,000 lines, and since january a hospital executive has to sign that it's accurate.
Astra opens it, finds what the hospital listed for your insurance plan, and reads your bill against it line by line. two numbers don't match, it flags the line and writes the dispute letter
mismatch found. one letter with both numbers in it. no lawyer
$2,300 ER visit in dallas, the file listed $890 for the same codes and the same insurer. $1,410 back. $18,900 for two nights in tampa, listed at $9,700. $9,200 back
her aunt does hospital billing and told her over christmas: 8 in 10 bills have a mistake and it's almost never in your favor. she has checked every bill in her own family for 15 years and never mentioned it outside the family
she tried by hand in january. one weekend on her own $1,900 ankle bill, 4 errors, $740 refunded. then she quit, the file is too big for a human. that excuse expired on saturday
she borrowed her roommate's pro account, dropped the same bill in and typed: "compare every line of this bill to the hospital's published price file for my insurer and write the dispute letter". same 4 errors. 6 minutes
she posted one line in the dorm group chat: send me your hospital bill, i keep 25% of whatever comes back, nothing if nothing does. 312 bills by monday. Astra cleared all of them in two nights. she has one group chat and a borrowed account. there are 6,000 hospitals
today her roommate saw the stripe tab and asked what his account has been doing all weekend. she showed him the numbers. he didn't say anything. he went and dug out his own ER bill from last year
$376,840 in letters out. the first 3 hospitals already corrected their bills, $4,100 off what people owed. 25% is hers
312 bills. 287 had a mistake in the hospital's favor
yours probably does too
Moved my AI agent to OpenClaw 2.0 tonight.
Four releases, one jump, a one-way migration in the middle.
Machine time: 25 seconds. My time: 70 minutes. Rolled back: nothing.
The upgrade was easy. The part that switched itself on was not.
thecooperlogs.substack.com/p…





















