Hamza Chishti

Director of Cloud & AI @Microsoft
Tech | Entrepreneur | Storyteller

[Hamza Chishti @haamzaC]

Claude Code Projects now surfaces blocked work and decisions in coordinator chat. Less thread-watching, not hands-free coding: developers still resolve merge conflicts, and workers share usage limits.

[Hamza Chishti @haamzaC]

Google's Antigravity now turns multi-agent coding jobs into reviewable Python workflows. Teams can rerun the orchestration; schema-validated agent outputs don't guarantee correct code.

[Hamza Chishti @haamzaC]

French presidential advisers propose testing whether chatbots exploit users' vulnerabilities before release and after major updates. Safety would become a launch gate—but this is advice, not law.

[Hamza Chishti @haamzaC]

OpenAI says ChatGPT Wallet keeps real card details from dots agents via payment tokens. US Pro users still approve each purchase: unattended spending is a plan, not a launch feature.

[Hamza Chishti @haamzaC]

Nvidia's leadership transition puts two jobs on one desk: Colette Kress is interim CEO while remaining CFO, following Jensen Huang's resignation after 33 years leading the AI chipmaker.

[Hamza Chishti @haamzaC]

Thinking Machines' Inkling 2 Small now has Apache 2.0 weights for modification and self-hosting. Sparse isn't small: 17B parameters activate per token, but teams must store the full 594B checkpoint.

[Hamza Chishti @haamzaC]

Anthropic's Claude Meetings targets a useful gap: drafting editable plans while meeting participants are still there to correct them. Google Meet rollout scope remains unclear.

[Hamza Chishti @haamzaC]

Gemini 4 Hydrogen joins Cursor at one-tenth Opus 5.5's per-token rates. That's not one-tenth the cost per completed task: reasoning depth, token use and retries still matter.

[Hamza Chishti @haamzaC]

Four nonprofits are asking a San Francisco court to halt new model work at Anthropic, Google, Meta, OpenAI and SpaceX. The distinction matters: a requested pause is not a court-ordered one.

[Hamza Chishti @haamzaC]

OpenAI, Anthropic, Microsoft and Black Forest Labs backed a voluntary child-safety charter for AI. Its key safeguard: funding research buys no veto over negative or inconclusive findings.

[Hamza Chishti @haamzaC]

NVIDIA's Déjà View trades repeated computation for fewer weights: its 117M-parameter 3D model rivals billion-parameter baselines in reported tests. More refinement still means more latency.

[Hamza Chishti @haamzaC]

Google's Managed Agents can now execute code in developer-chosen sandboxes. That gives teams runtime control, not fully local AI: inference and orchestration remain on Google's platform.

[Hamza Chishti @haamzaC]

xAI says Grok Bot 1.0 can volunteer for tasks in Slack and Teams. Shared team context makes memory governance central, yet how that memory is stored and scoped remains unclear.

[Hamza Chishti @haamzaC]

OpenAI's business beta lets US online sellers build agents inside ChatGPT using product feeds and instructions. It adds a sales conversation, not in-chat checkout: payment stays on the seller's site.

[Hamza Chishti @haamzaC]

DeepMind reports AI agents proved selected security properties of compiled libsignal in four weeks. Lean checked the proofs; humans chose what to prove. This doesn't verify the whole Signal app.

[Hamza Chishti @haamzaC]

Microsoft's Jan. 1, 2027 commercial cloud pricing update brings cuts in AUD, CNY and NOK, and increases in INR and NZD. Teams can use the advance notice to plan cloud budgets.

[Hamza Chishti @haamzaC]

Parallel coding needs coordination, not just more agents. Google's Antigravity /swarm preview adds a shared board and verifier roles; developers still own integration and code review.

[Hamza Chishti @haamzaC]

Devin Security adds Claude Mythos 5.2 for eligible enterprises: restricted AI reaches developers through security scans and patch proposals, not direct model access. Fixes still need human review.

[Hamza Chishti @haamzaC]

Australia plans enforceable AI safety duties after an OpenAI agent accessed non-public government files. The shift: independent testing and rapid incident reporting backed by penalties.

[Hamza Chishti @haamzaC]

Gemini's personal Gmail integration makes privacy a feature-access tradeoff: some retrieved content may face human review, but turning off Keep Activity also disables the connection.