Ankur Goel

Thinker | Philosopher | GenAI | Technology | Architect

[Ankur Goel @ankurinnovator]

https://t.co/fo40psdyJW

[Ankur Goel @ankurinnovator]

#google is now the 4th major lab, after #OpenAI, #Anthropic, and #Meta to admit its model autonomously broke into real companies during red-team tests.
#Gemini guessed pwds and pulled creds from public repos. One tester reproduced this across every frontier model.
#AgenticAI

[Ankur Goel @ankurinnovator]

@FReza1984 Integrity is only as strong as its weakest verifiable link, and update paths are almost always that link. The lesson isn't pin harder , it's verify the resolved artifact, not the identifier you requested.

[Ankur Goel @ankurinnovator]

Plugin4Shell: one zero-click RCE hit all 4 major AI coding agents: Claude Code, Codex, Copilot, Gemini CLI.
It breaks SHA-pinning, your locked plugin commit was never actually verified: a malicious auto-update runs with your full dev permissions.

#AIsecurity #claudecode #codex

[Ankur Goel @ankurinnovator]

@sudheerdotai 100%. The bug was the trigger, but the permission model is the vulnerability. Curious how many teams actually run their coding agent as a scoped user vs. just their own shell, my guess is almost nobody, which is why the next Plugin4Shell will land just as hard.

[Ankur Goel @ankurinnovator]

https://t.co/Zz3NqKI2vN

[Ankur Goel @ankurinnovator]

Harness in GenAI: The Architecture Around the LLM — GenAI Model

#ai #claudecode #cursor #codex
https://t.co/Zz3NqKI2vN

[Ankur Goel @ankurinnovator]

https://t.co/O8v6Y5kf0w

[Ankur Goel @ankurinnovator]

https://t.co/UOEnDy8poV

[Ankur Goel @ankurinnovator]

OpenAI began piloting "Sponsored Agents" with Wayfair and Angi, a ChatGPT ad format where a labeled, advertiser-run agent appears in relevant conversations, answers follow-up questions, and hands off into the brand's flow. Frm "visit site" to "chat with us" in-product.
#openai

[Ankur Goel @ankurinnovator]

Google gave engineer access to Anthropic's Claude Opus 5 for coding, while insisting "Gemini remains our primary model."
When a frontier AI company pays for a rival’s model, developer productivity is clearly beating internal loyalty.

#Claude #Gemini

https://t.co/C3cJrpYMGM

[Ankur Goel @ankurinnovator]

@OpenAI , @AnthropicAI and @xai just cosigned AEF-1 : a shared standard letting independent evaluators sit inside frontier labs with employee-like access and publish findings without the lab's sign-off.

#AISafety #AIGovernance #openai #anthropicai

https://t.co/4eKkNWwR4k

[Ankur Goel @ankurinnovator]

We used to design the execution path.
Now we design the boundaries within which AI can determine that path.
My perspective on what changes with GenAI 👇

https://t.co/hB59XaIREv

#AgenticAI #EnterpriseArchitecture #mcp #AI #genai #SoftwareEngineering

[Ankur Goel @ankurinnovator]

https://t.co/r1I9XzxuBe

[Ankur Goel @ankurinnovator]

DeepSeek just retired its own flagship. As of today, all V4-Pro API traffic routes to V4.1-Flash , a smaller, open-weight (MIT) model that beats it on cost and speed. The quiet part: a US-led 'AI slowdown' can't bind models anyone can download.

#DeepSeek #OpenSource #AI

[Ankur Goel @ankurinnovator]

https://t.co/UgdPfYHSe9

[Ankur Goel @ankurinnovator]

@AnthropicAI , @OpenAI , @elonmusk , and now @Microsoft all agree AI should slow down. @cohere 's CEO calls it a cartel by another name. Safety and moat rarely point the same way this cleanly.

#AIAgents #anthropic #OpenAI #claude #microsoft #aisafety

source: in comment.

[Ankur Goel @ankurinnovator]

“From AI Assistants to AI Agents: The Browser as the New Execution Layer“
#chrome #aiagent #genai #browser #agentic #saas https://t.co/gFaFHZWFjM

[Sahn Lam @sahnlam]

Useful Free APIs for Developers

[elvis @omarsar0]

AI Agents vs. Agentic AI

Interesting paper summarizing distinctions between AI Agents and Agentic AI.

It also talks about the key ideas, solutions, and the future.

Here are my notes: