Archit Dwivedi

Home

Hi, I’m Archit

Notes on building AI agents that ship

Essays on agents, context engineering, and the runtime around the model. Written from the production side. New posts most weeks. More about me · what I’m working on.

Timeline

  • AI updates of 17 May, 2026
    NVIDIA open-sources a frontier video model that runs on one GPU
  • AI updates of 14 May, 2026
    Google folds Gemini into Android; defense AI hits $61B
  • AI updates of 11 May, 2026
    OpenAI opens its self-serve ChatGPT Ads Manager to every US business, marking the moment the most-used AI product on the consumer internet becomes an ad-supported platform — while Anthropic teaches Claude agents to dream, NVIDIA earmarks 5 GW with IREN, and the EU softens its AI Act.
  • AI updates of 9 May, 2026
    Cloudflare cuts 1,100 jobs and points the finger at AI — the moment displacement leaves the abstract. Plus Anthropic teaches Claude agents to ‘dream’ between sessions, ChatGPT swaps in GPT-5.5 Instant, and an AWS US-East outage knocks Coinbase and FanDuel offline.
  • AI updates of 8 May, 2026
    DeepMind says AlphaEvolve, its Gemini-powered coding agent, is now optimizing real systems inside Google — DNA pipelines, power grids, even next-gen TPUs — while Anthropic teaches Claude agents to “dream” between sessions.
  • Hands on with OpenCLI
    OpenCLI wraps existing CLIs and websites into one hub your coding agent invokes through a single skill file. Stateless, deterministic, and a real alternative to MCP servers when token cost matters.
  • AI updates of 7 May, 2026
    Anthropic locks all of SpaceX’s Colossus 1 (220k+ GPUs) and doubles Claude Code rate limits the same day. Cloudflare agents start buying domains. Coding-CLI cluster ships: Claude Code 2.1.132, Codex 0.129 /goal, Copilot CLI 1.0.41, Cursor 3.3, OpenCode 1.14.40.
  • Hands on with NVIDIA OpenShell
    Install, sandbox an agent, watch it get blocked, then selectively unblock it. A hands-on walkthrough of NVIDIA’s policy-driven sandbox runtime for autonomous AI agents.
  • AI updates of 6 May, 2026
    OpenAI quietly flips ChatGPT’s default to GPT-5.5 Instant; Google ships free 3× Gemma 4 inference via MTP drafters; Sierra hits a $15B valuation. Plus Claude Code 2.1.128, Codex /goal, Cursor SDK, and a GitHub MCP secret-scan GA.
  • AI updates of 5 May, 2026
    OpenAI’s GPT-5.5 lands with native Skills, MCP, hosted shell and computer use — adopting Anthropic’s agent stack wholesale, while Cursor opens its SDK and Claude Code 2.1.128 ships zip plugins.
  • AI updates of 4 May, 2026
    Anthropic flips its repo-vulnerability scanner — Claude Security — into public beta on Opus 4.7. Codex CLI 0.128 ships persisted /goal flows alongside a 1000-tok/s GPT-5.3-Codex-Spark research preview, while Cursor adds Security Reviewer + Vulnerability Scanner agents.
  • Claude Security hits beta; Codex, Cursor, Grok ship agent upgrades
    Anthropic ships Claude Security in public beta — code-vuln scanning becomes a real product. Plus: Codex CLI persisted /goal, Cursor Security Review, Grok 4.3 with 1M context.
  • Pentagon picks 8 AI vendors without Anthropic; Cursor SDK ships
    The Pentagon greenlit eight tech firms to run AI on its classified networks and pointedly excluded Anthropic; meanwhile Cursor turns its agent runtime into a TypeScript SDK and Claude Code 2.1.126 ships project-purge tooling.
  • GPT-5.5 ships with hosted shell; OpenAI breaks Azure lock-in to AWS
    OpenAI launched GPT-5.5 with a 1M-token context, hosted shell, and native MCP — and within 48 hours AWS Bedrock began hosting OpenAI products, ending five years of Azure exclusivity.
  • Prompt → Context → Harness: three layers of LLM engineering
    The same task gets bigger and more system-shaped as you climb. A hands-on tour of the three layers — with code, diagrams, and the concrete differences between them.
  • Cursor opens its SDK, ships security agents; Codex CLI lands /goal workflows
    Cursor pivots from IDE to agent platform in a single week — TypeScript SDK in public beta, Security Reviewer agents that comment on every PR, and a Team Marketplace — while OpenAI’s Codex CLI 0.128 ships persistent /goal workflows and Claude Code v2.1.126 lands project purge.
  • Cursor SDK lands; Codex, Claude Code, OpenCode all ship in one week
    Cursor opens its agent runtime to everyone with a public TypeScript SDK; Codex CLI ships /goal workflows, Claude Code adds PostToolUse output rewriting, and OpenCode keeps shipping — coding agents become a programmable substrate this week.
  • Welcome back — what this blog is now
    After a year of building agents instead of writing about them, I’m starting again — with a different shape. Less tutorial, more second-order.
  • AI AWS DEVOPS
  • AI Trip Planner
    Introducing the latest innovation in my AI assistant series: the Travel Planner Assistant. This assistant is a game-changer for anyone looking to travel, whether… Read more: AI Trip Planner
  • AI Cricket Bot(IPL)
    In the evolution of my AI agent creations, I’m excited to introduce AI Cricket Bot in the series, specifically tailored for cricket enthusiasts. This… Read more: AI Cricket Bot(IPL)
  • AI Developer
    In my last post, I discussed how generative AI has the power to revolutionise the typical duties of a Project Manager by automating complex… Read more: AI Developer
  • AI Project Manager
    Revolutionizing Project Management: The Rise of AI Project Managers As a developer, my firsthand experience with an AI project manager that leverages Google Sheets… Read more: AI Project Manager
  • AI Part 2: Chain the A.I
    AI Chain In our last blog, we delved into the world of Large Language Models (LLM). Moving forward, this post will explore the concept… Read more: AI Part 2: Chain the A.I
  • AI Part 1: Everything & yet nothing about LLM
    What is Large Language Model? It’s essentially an advanced form of generative AI, specifically designed to understand and generate human language, trained on vast… Read more: AI Part 1: Everything & yet nothing about LLM

— say hi at /contact