UC Berkeley's Rubin Lab reportedly got results 15x sooner on a DGX Spark by avoiding an 11-hour shared-cluster queue.
That's more than a working day waiting for computation to start.
NVIDIA's new DGX Spark HPC Developer Kit connects four desktop systems in a ring, with 512GB of combined memory, 20TB of storage and no network switch required. It supports models up to 700B parameters using NVFP4 quantization.
Local AI deserves a place in company infrastructure plans for exactly this reason. A team needs room to test an idea, find the mistake and run it again. Waiting for shared compute delays each attempt.
For companies with research teams, I'd divide the work accordingly. Develop and debug locally, keep proprietary datasets on premises, and reserve shared GPU capacity for production runs.
The kit is on sale. The HPC SDK installation guide is still coming soon, so I'd account for that when planning how researchers will get the software running.
I'd start with the workloads where people lose the most time waiting for compute, then compare the full local cost with the cost of those delays, including the experiments a team never gets around to running.
Include waiting time in the compute budget.
Cigna Healthcare case managers used to click through more than 10 separate systems just to get ready for a member call.
That's the everyday work OpenAI was pointing to yesterday, when it showed computer-use agents moving through the software healthcare teams already use. The agents work in the existing screens, clicking, typing, and copying the way a person would.
Cigna's CIO, Madhu Nutakki, said they went from concept to deployment in under eight weeks. The agent can now pull together benefits, coverage, providers, appointments, and prior interactions, so a case manager spends less time hunting and more time with the member.
CVS Health is further along in production. One agent cleared a months-long document-sending backlog over a weekend. Another handles the system steps for tens of thousands of specialist consults each day. CVS says those agents run with validation checks and human oversight, including screenshots and a record of each action.
OpenAI also published its own evaluation numbers. Reviewers rated 99% of 325 EHR inbox tasks as safe, and physicians called the work "very helpful" or "extremely helpful" in 93% of 646 assessments. Those are OpenAI's figures, on tasks it designed with physician input.
Less time spent hopping between screens means more time with members. But these agents are being asked to enter portals, draft orders, and prepare authorization requests in software that was never built for them. The organizations doing this carefully are putting a person in front of the submit button, with a log of what the agent saw and did.
If you lead a team that's drowning in handoffs between systems, this is the version of AI that will show up at work: something that can operate the tools your people already live in.
So before you turn one loose, pick the person who watches the last click.
ChatGPT for Healthcare is advancing with more capable computer-use agents, stronger clinical answers, and 20+ new healthcare plugins.
@AbridgeHQ, @AmbienceAI, @Cigna, @CVSHealth, and other leading healthcare organizations are putting our models to work across prior authorizations, referrals, discharge planning, and patient outreach.
Together, these advances connect medical evidence with practical help across the tools teams already use—so they can spend less time navigating systems and more time caring for patients.
https://t.co/NI35X8pG57
Some of Steve Yegge's players have been in his online game since 2001. Last week they asked him for something unusual: slow down.
His AI agents were shipping 50 to 100 new features a day. The players liked the new stuff. The pace scared them, so they asked for a roadmap.
That's from his talk at AI Engineer CODE, and the rest of his list sounds familiar. His agents flooded his merge queue with tickets and overloaded his machines, with only about a dozen people involved.
His advice to CIOs and CTOs now: slow down. Pace AI to human speed.
Most people will read that as a story about coding agents. I think it's about every team rolling out AI right now.
Output isn't the hard part anymore. The hard part is everything after it: the person who has to review the draft, approve the change, answer the customer, or reconcile the invoice.
If AI doubles what lands on their desk and nothing else changes, your team didn't get faster. You just moved the pile.
So before you add another agent, answer three questions:
Who reads what it produces?
How much can that person actually absorb in a week?
What happens to the work they don't get to?
Set the pace by that answer, not by what the tool can do.
Yegge's players weren't against AI. They just needed to keep up. Your team is the same.
I just finished talks about my software factories at two conferences. Afterwards a bunch of people came up and told me what they were surprised about. Here is a distillation of what people thought were hot takes from my two talks.
First hot take: I tell everyone to slow down. Most speakers go on about how fast they're going, but I strongly believe in pacing AI outputs to human speeds. I regularly tell AI-pilled CIOs and CTOs to chill: almost nobody is _actually_ going fast with AI. Engineering still moves at engineering speeds, and business moves even slower. The ones pushing too hard are starting to see downtime for it.
If you don't pace your AI to human speeds, everything starts breaking around you, even at small operations like my online video game. I only have about a dozen humans working with my factory, and only a few hundred active old-timer players in the community who get along well, and still, introducing AI has still been very rocky.
My agents have, at times, overwhelmed my players with features, my accountant with emails, my merge queue with tickets, and my machines with load. Agents have pissed off my admins, resulting in name-calling, they've made messes in the game, and they've launched broken OSS without permission. The list goes on.
I can get away with all this chaos because everyone involved in my operation is very forgiving. They're excited by the potential. But even so, I had to learn to slow down, real fast. And so that's what I preach.
Hot take #2: Should you buy a software factory? Do you even need one? No. You probably don't need one. Factories take an extraordinary amount of upkeep. They turn into garbage mountains unless you put a frightening amount of time and tokens into curating them. I had to burn my previous five factories down since it eventually became cheaper to rewrite them than to fix them. So I'm not out there advocating for software factories or swarms just yet. If you do want one, be aware that it is a significant commitment of both money and human leadership to keep one running.
Several people told me that I was the only one who stood up onstage, at either conference, and said Build, don't Buy. But my experience is that software factories grow to be highly bespoke to your domain. Mine is super tangled up with my game, and I think that's inevitable. Because the factory is not just building my product, it's supporting it too. It's a company in a box. And they're easy for Claude to build for you.
Third and arguably hottest take: Next year will have a lot of human/agent and agent/agent office drama, resulting from bringing in AI employees, and it will be gloriously funny, but also frustrating.
Many of you are already on this bandwagon, but everyone next year will have a personal chief-of-staff agent. Busy people, even hobbyists, will have whole entourages of helper agents. The hot take is that people will become so dependent on these little private entourages, that they'll bring them to work. Your people will talk to my people, etc. Whole protocols will emerge for private entourages negotiating work between humans, inside companies big and small.
Nearly all humans will wind up having to start working with AI employees next year, and it'll be the beginning of the biggest cultural change we've seen in a hundred years.
So that's the hot takes. Slow down. Only build a factory if you think you really need one. And focus on empowering yourself with your own agentic chief of staff (or team). Your entourage will define how you work with others, so start by perfecting how you work with them yourself.
Microsoft 365 email went down across more than 70 countries for about 30 hours. Microsoft has restored service after at least its fourth significant outage this year.
In January, Microsoft reported more than 450 million paid commercial seats. Pennsylvania lawmakers went about 18 hours without email during this outage. Securities firms and asset managers in Japan reportedly turned to fax machines when trade-confirmation emails stopped arriving.
An inbox outage meant getting the fax machine working again to confirm trades.
Companies building AI into their operations need to take this dependency much more seriously. An agent handling work through Microsoft 365 depends on that infrastructure being available. Giving it more responsibility increases how much work depends on the same connection.
Recovery deserves as much attention as the outage. Simultaneous reconnection attempts made restoration harder, and delayed email was still moving through queues as service returned.
For a company automating work from its inbox, that backlog needs a deliberate recovery process. Decide which delayed requests are still valid, which actions need fresh approval, and how to prevent retries from doing the same work twice.
Test what happens when the inbox comes back.
Claude shipped Personal Finance in the US today. Connect the accounts you choose, ask about your own balances and transactions. Access is read-only. It can't move money or invest.
Introducing Personal Finance in Claude, in beta in the US.
Connect the accounts you choose, and Claude answers your money questions in everyday language, from your own balances and transactions. Access is read-only, and Claude can't move, manage, or invest your money.
@ajambrosino@davis7 Please get rid of the project view and let chat be more like an inbox ! The project view is so hard to navigate with all these agents