vikas mittal

@robinebers looks super useful! Tired of constantly refreshing feeds for new model drops 😅 Would love to join the beta and test it out. Any chance for a TestFlight link? Thanks for building this, Robin! 🚀

fuck it, I'll keep it free I built an app that notifies you within seconds of a new AI model being released so I can stop refreshing my feed every minute 🔴 LIMITED BETA 🔴 → 100 TestFlight spots → only on iPhone / iOS 26+ → completely free while in beta will add more features if people actually like this link in first comment 👇🏻

your first misconfigured tool permission is the one that haunts you. your first agent loop that wouldn't stop. your first log that lied to you. you just have to make sure that first failure… doesn't make you stop shipping…and become your last deploy.

the highest-leverage skill in network security right now is agent design. if you want to build safer, more reliable AI agents, here are the resources that shaped how I think about it: 1/ The Practice of Network Security Monitoring - Richard B. 2/ Threat Modeling - Adam S. 3/ Building Secure and Reliable Systems - Heather A. 4/ The Art of Invisibility - Kevin M. 5/ Designing Distributed Systems - Brendan B. 6/ Accelerate - Nicole F. 7/ The Phoenix Project - Gene K. 8/ Staff Engineer - Will L. read these, then go ship something small and local. what would you add to this list?

#UCLFinal is a reminder for AI agents too: flashy attacks get attention, but systems win on structure. For security agents, that means clear roles, least privilege, logged actions, and rollback paths. Autonomy needs defense as much as offense.

Building AI agents for security has a truth: the agent is not the product. The product is the control plane around it: least privilege, replayable logs, approval gates, rollback. Autonomy only works when the blast radius is visible. Where packets meet prompts.

Karpathy joining Anthropic is not just talent-war news My read: frontier labs now need people who can turn deep research into clear builder intuition The edge is not only better models It is better explanations that help engineers use them safely

Gemini Omni is the Google I/O trend I am watching: multimodal AI is moving from "make a clip" to "reason about what should happen next" Security edge: once models simulate actions, we need packet-capture style logs for agent behavior What would you log first?

Gemini Omni is the Google I/O trend I am watching: multimodal AI is moving from "make a clip" to "reason about what should happen next." Security edge: once models simulate actions, we need packet-capture style logs for agent behavior. What would you log first?

Codex limit resets are not just a perk. They are a signal that agents are moving from human-paced clicks to machine-paced work loops. The real product layer now: quotas, retries, checkpoints, cost controls, and clear blast radius. QoS for agents.

Google I/O signal is simple: Gemini is moving from chat box to control plane across Search, apps, devices, and coding. Network-security angle: every AI agent needs the boring stack too: identity, least privilege, audit logs, rollback, and rate limits. Packets meet prompts.

AI agents are less like chatbots and more like service accounts with reasoning. That changes the security model. The question is no longer only: did the prompt look safe? It is: which tools could the agent reach, what did it mutate, and can we replay the decision trail?

Hot take: AI agent pricing is really a reliability debate. Unlimited plans trained builders to run agents like noisy broadcast traffic. Metered credits force better architecture: smaller tasks, checkpoints, retries, logs, and clear blast radius. QoS for prompts.

Runtime AI security is the new blind spot. DLP covered stored data. CASB covered SaaS traffic. EDR covered endpoints. But agents read, reason, call tools, and mutate state. That needs policy, audit trails, least privilege, and kill switches.

AI coding agents are moving from autocomplete to remote ops. Codex on mobile is the signal: start work anywhere, keep it in a bounded host, review the diff later. For network engineers, this is SSH jumpbox logic wearing an AI hoodie.

@hqmank Cursor

🎁 Cursor 50% off Half off your first month of Pro, Pro+, or Ultra. I have a few invite links left. Reply "Cursor" if you need one. I’ll send them while I still have invites.

@cognition @cerebras Would love to try Max for terminal-heavy security engineering: prototyping detection scripts, reviewing infra changes, and cutting through noisy logs faster. SWE-1.6 Fast + Cerebras at 1000 tok/s sounds perfect for real ops work.

Intelligence at 1000 tokens per second, right in your terminal. Now available with SWE-1.6 Fast, powered by @cerebras. We're giving the first 100 people who respond a free month of Max to try it out.

@himanshutwtxs @theomkute @LocalDev_ Codex

@HeyGen CLI

Your AI agent can now generate and ship videos. HeyGen CLI is now live. Run one command and your agent handles it all: script → avatar creation → video → delivery All from the terminal. Just your agent and the CLI. RT + Comment “CLI” and we’ll DM API credits (must follow)

@Pokee_AI PokeeClaw

OpenClaw doesn't belong in production. We built PokeeClaw — enterprise-secure AI agents, zero setup, 1,000+ app integrations. Try now: pokee.ai First 500 to follow @Pokee_AI, comment “PokeeClaw”, like & repost get 1 month free.

@itsumeshk @runclaw let's give it a try

We're launching RunClaw to kill OpenClaw. OpenClaw costs $700 to set up. RunClaw costs $1 no setup. > OpenClaw can’t build you a website > Can’t generate a video > Can’t make a slide deck > Has 9 security CVEs RunClaw does all of it. Better agents. More secure. Always in your DMs. Try it now for $1.