Microsoft Ignite 2026 is live, and the big theme is AI agents moving straight into Teams.
Here's what they just announced:
1. Agentic Users (public preview)
Agents can join a Teams conversation as participants.
2. Agent 365 Business Analytics
Think of it as an ERP for AI usage. It ties AI spend to outcomes and ROI. Available through the Frontier program.
3. Perception (now GA)
Teams of red, blue, and green security agents guard your agent setup.
4. Copilot plugins
New plugins for Dynamics 365, plus a Salesforce integration coming to Copilot.
5. Meta's latest models are coming to Foundry.
Satya opened by owning last week's Outlook outage: "We have work to do."
Copilot is being pitched as the new OS for work.
59
Satyabrata Dash retweeted
Google believes it has found the first cyberattack run end to end by AI.
A framework compared software before and after each patch, worked out the bug, wrote the exploit, and launched it. No human in the loop.
Here's what it means for anyone in Tech 🧵(1/9)
Meditate Madi
Rameshwaram Cafe, like you have never seen before 🤍
The rush outside. The stillness within.
7 minutes, eyes closed, right in the middle of the buzz.
Many minds. One silence. 🧘
Every day, people of every age pause at The Rameshwaram Cafe, Bommasandra, for Miracle of Mind, Sadhguru's free 7-minute guided meditation.
✨ Free sessions every morning
✨ No experience needed
✨ All are welcome
Join us every morning and make your Mind a Miracle in 7 minutes. 🙏
📍 The Rameshwaram Cafe, Bommasandra, Hosur Road, Bengaluru
📲 Download the free Miracle of Mind app
#MiracleOfMind #RameshwaramCafe #Bommasandra #Bengaluru #Meditation
6
MSD retires from the IPL as a player. He will be a mentor for the CSK.
A dream career for Thala.
2
34
A vendor got breached, and the damage reached its customers' customers.
Two stories from this month show the same pattern.
First, Fireworks AI. The AI infrastructure company says an attacker got into its development environment through a third-party cloud provider, and that billing data, resource metadata, and for some customers, datasets, secrets and incomplete model weights were exposed. Notion, which used Fireworks to run part of its AI features, says transcripts of Notion AI chats from under 1% of its workspaces were caught in it. Notion has told affected users to rotate any passwords or API keys they pasted into those chats.
Second, Qubole. Security researchers at Wiz say an attacker used the AWS credentials of Qubole's data platform to assume roles in multiple customer AWS accounts between October 14 and 26, and stole secrets and data. Nobody needed to break into each customer one by one. The trust those customers had already set up with Qubole was the way in.
Why this is becoming a bigger deal:
Companies now plug in dozens of AI tools, data platforms and integrations. Each one holds keys, tokens or access to your data. Your security is only as strong as the weakest vendor you trust.
AI makes it worse. Prompts, chat history, datasets and API keys often end up stored with third-party model providers that most users never see.
Some attackers are not even after the model itself. Fireworks' CEO said the apparent motive was not stealing model weights.
What to do this week, if you run anything on these tools:
Rotate any keys or passwords that were ever pasted into an AI chat.
Audit cross-account roles and remove trust relationships you no longer use.
Ask your AI and SaaS vendors where your data is stored and who else processes it.
Vendors are now part of your attack surface. Treat them that way.
1
20
There were and still are Yogis before science started exploring the architecture.
This is called Trataka in Yoga. There are many variations of this. The studies were already done in India proving efficacy in improving concentration years ago before this 2020 western research paper quoted here.
This method will also become a trend in the west without mentioning Trataka and its Yogic roots. Like many Pranayama and Yogic methods are now rebranded as this guy's method or that man's technique.
What we have to understand is, yogic methods, paranayama techniques, herbal medicines mentioned in Indian texts thousand plus years ago are first studied in India and research papers published. Indians themselves tend to think this is not the case. They are wrong.
These India done studies never get recognition or attention. Then some lab in the west takes cognizance of them or a new trend emerging from such ancient practices, does its own experiments proving efficacy. Then it immediately becomes "scientific" and gets rebranded and marketed well by the west.
The best example is Turmeric, which was tried to be patented in the 90s after western research proved it's antibacterial, anti-inflammatory properties. But those were figured out by Indian scientists decades ago with their own scientific research. And more importantly figured out by ancient Indians tacitly and used as such for millennia.
My main issue with this is dilution of ancient knowledge is, be it Yoga Nidra (Which I must say Huberman did promote giving due credit after backlash), or Ayurvedic herbs, or breathing methods, the rebranded modern versions are pale improper versions of the original.
If the original attribution is preserved, then science can slowly research a lot more aspects and techniques of these ancient practices to create much more robust scientifically validated modern methods to help a lot more people based on their specific needs.
But by trying to hide the original attribution, and making it mainstream as some new scientific method branded under some western guy's name, great injustice is done in taking the full and proper knowledge found out thousands of years ago by ancient cultures to the world.
2
46
Satyabrata Dash retweeted
Claude writes better code when it can verify its work. Checks that used to be expensive, like tests, benchmarks, and end-to-end runs, are now cheap.
Here’s how the Claude Code team thinks about verification:
claude.dev/blog/how-the-clau…
Traditional software engineering and AI-assisted engineering look different on the surface.
Underneath, most of the steps still line up.
🧵
1
16
Incident response becomes rollback and kill switches.
When an agent goes wrong, the fix isn't always a hotfix.
Sometimes it's pulling its permissions, reverting its changes, and replaying the session to see where the reasoning slipped.
1
3










