7 October, 19:00 CEST: @Windows, @surface and @nvidia live on one stage.
Public so far: "a look at what's next", including RTX Spark. Pavan Davuluri, in May: "Great silicon deserves deep platform work."
Large AI models, run locally on a PC. I'm in.
https://t.co/PLNJhElZKo
Sales reviews have an agenda item nobody writes down: whose number is right.
Jessica Hawk: with Fabric IQ, Copilot answers from the same Power BI model as the reports.
Only works if the definitions are agreed first.
https://t.co/aJDyr4btpZ
"No model is best at everything. They're trained differently, so they fail differently." (@jared_spataro, @Microsoft, 23 Sept)
Category logic applies: you don't let one supplier own the whole shelf. Ask a tool for its failure profile, not its benchmark.
Copilot Autopilot (formerly Scout): "You set the objective and boundaries; Autopilot handles the rest."
That is delegation, defined. If a team can't write objective and boundaries down for an agent, it never did for people either.
https://t.co/lC74829Oe5
The new Surface Pro 12 inch and Surface Laptop 13 inch bring Snapdragon X2 Plus to the same lightweight, fanless designs.
What I like: more capability without giving up portability. In retail, I’d lead with the experience people can try themselves.
https://t.co/TmLopDDL7o
A practical CCS example: AI helping prepare Surface performance analysis and commentary.
The value I care about is better decisions with our partners, not just faster reports. People still own the judgment and the action.
https://t.co/D8hSZX3yX2
For AI study help on a PC, I'd rather get a hint and make another attempt before seeing the worked answer (tempting though it is). Microsoft's education piece makes a useful case for keeping the learner involved in the thinking.
https://t.co/kmyhavtrDI
When showing consumer Copilot on a Surface to a family, I'd explain the age and sign in requirements first. A demo on an adult account leaves that question open (even if the answer looks good).
Microsoft's current guidance:
https://t.co/JOjlwiQyg9
Microsoft's documentation confirms Task Manager can now show NPU resource usage.
Run an AI enabled feature and watch whether that graph moves. It is a quick first check, not a full trace, but it beats taking the spec sheet on faith.
https://t.co/W3xEmfILKM
6/
It breaks in one place. It cannot see the meeting where I sit quiet for fifty minutes and settle the whole thing in the last four, which from outside looks exactly like a dead one.
So LEAVE is a list worth checking, not instructions. I have overruled it more than once.
1/
I have three Cowork prompts running: what came in overnight, what I owe people by tonight, what the week added up to.
None of them touched the meetings themselves, which is the part that actually eats the day.
So here is the fourth. Monthly, not daily.
5/
Grade each series KEEP, SHORTEN, DELEGATE or LEAVE with one sentence of reason. LEAVE only where I did not speak and nothing came out of it.
Then draft the message to the organiser, in my voice, under five sentences, and never make the reason about being busy.
🌐 How is AI transforming network operations at Microsoft? With Network Infrastructure Copilot (NiC) and AIOps, we've saved 20,000+ hours on network management, reduced engineer toil, and improved reliability. See what we've learned as Customer Zero. https://t.co/QkUSM1r0mv
Better teamwork often comes from fewer interruptions.
I appreciate how our approach to collaborative agents in Microsoft Teams helps people stay informed without adding more noise: https://t.co/NzVoyTXJzn
I used to measure an AI rollout by adoption. Who's logging in, how often.
That's not progress, that's attendance.
Now I ask which piece of work it replaces, and who still owns it in month three.
That last one kills more tools than the rest, which I've made my peace with.
Encryption standards face a future challenge from quantum computing, putting long-term data security at risk.
We’re advancing quantum-safe algorithms and tools to prepare now.
Read BleepingComputer’s coverage: https://t.co/DBHxe59f9v