AI Pulse #1 — New Frontier Models, MCP Goes Vendor-Neutral, and Video Eats E-commerce

    This is the first edition of AI Pulse — a short, curated digest of what actually matters in AI right now, with my take as someone shipping AI products daily. No hype recycling: each item is something I'd bring up in a product planning meeting.

    In this edition
    1. 01 June 2026

      Anthropic launches Claude Fable 5 (and Mythos 5 for trusted access)

      Anthropic released Claude Fable 5, its most capable generally available model — state-of-the-art on most published benchmarks, with notable gains in software engineering, knowledge work, vision, and long-context performance. A second variant, Claude Mythos 5, ships without some dual-use safety measures to approved organizations only.

      My take

      The two-tier release (general vs. trusted access) is the interesting product decision here, not the benchmarks. If frontier vendors normalize capability-gated tiers, AI product teams will need to think about which tier their workloads legally and practically qualify for — the same way we already think about data residency.

      Source: Anthropic Newsroom

    2. 02 May 19, 2026 (I/O)

      Google ships Gemini 3.5 Flash, with 3.5 Pro weeks away

      Gemini 3.5 Flash launched at Google I/O and is already the default model in the Gemini app and AI Mode in Search. Gemini 3.5 Pro was announced for June.

      My take

      Flash-tier models being 'good enough' as defaults is the real story for builders: the cheapest tier of each vendor keeps absorbing use cases that needed a mid-tier model six months ago. If you haven't re-evaluated your model routing since early 2026, you're probably overpaying.

      Source: CNBC

    3. 03 June 2, 2026

      Microsoft unveils MAI-Code-1-Flash, its first in-house coding model

      Microsoft announced MAI-Code-1-Flash, an in-house model that generates application and website source code from written descriptions — explicitly positioned to reduce reliance on OpenAI and lower costs for developers.

      My take

      Every platform that resells inference eventually builds its own models for the high-volume, low-margin tasks. Expect coding-agent pricing to keep falling — and expect 'which model wrote this' to matter less than the harness around it.

      Source: CNBC

    4. 04 June 2026

      MCP is now a Linux Foundation standard — 110M monthly downloads

      The Model Context Protocol, donated by Anthropic to the Linux Foundation's Agentic AI Foundation, has hit 110 million monthly downloads — adoption faster than React — with the official registry passing 9,400 servers, up 407% since its September 2025 launch.

      My take

      Vendor-neutral governance is what enterprises were waiting for. If you build a SaaS product and don't expose an MCP server yet, you're choosing not to be reachable by the fastest-growing client surface in software. This is the new 'do you have an API?'

      Source: Tech Hub Weekly AI Roundup

    5. 05 June 2026

      AWS MCP Server goes GA

      AWS made its managed remote MCP server generally available — authenticated access to AWS services for agents through a small fixed toolset, with IAM context keys for fine-grained permissions and reduced token usage for complex workflows.

      My take

      The design detail worth copying: a small, fixed set of tools instead of one tool per API. Tool sprawl is the silent killer of agent reliability — fewer, well-designed tools with good parameter schemas beat hundreds of thin wrappers every time.

      Source: AWS News Blog

    6. 06 June 3, 2026

      AI video becomes the default content engine for e-commerce

      Industry reporting marks 2026 as the turning point where text-to-video replaces production houses for brand content: product videos that cost $500–$2,000 with a crew now cost pennies, generated straight from the product data merchants already have.

      My take

      I build exactly this at Modelia, so I'm biased — but the shift I see from the inside is real: merchants stopped asking 'can AI make me an image?' and started asking 'can it make me the whole campaign?' The platforms that win will be the ones that connect generation to the merchant's actual catalog and publishing workflow, not the ones with the prettiest single output.

      Source: Digen (citing Cybernews)

    • Frontier Models
    • MCP
    • AI Agents
    • E-commerce AI

    Share

    LinkedInX
    Portrait of Harsh Rastogi, AI Product Engineer

    Harsh Rastogi

    AI Product Engineer and Business AI Head at Modelia

    I build AI products and agents from the first sketch to production at Modelia, and I'm a Founding Engineer at SelfAgentic. Before that I built platforms at Asynq and Bharat Electronics Limited. I publish the agent skills I use every day, and you can see what I've shipped in my work.

    Go deeper in the latest articles.

    • Claude Fable 5.2: 20% Cheaper, 30% Faster, and What Changes for Developers

      Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, 2026: $8 input and $40 output per million tokens (down from $10/$50), output up to 30% faster than 5.1, 55% fewer safety-classifier blocks, and big jumps on agentic benchmarks. Here's the pricing math, the benchmarks against Opus 5.5, the four breaking API changes, working TypeScript, and when Fable 5.2 is actually worth it over Opus 5.5.

      • Claude
      • Anthropic
      • Claude Fable 5.2
      • LLM Pricing
      AI & Machine Learning11 min read
    • What Is an AI Product Engineer? The Role, the Skills and What the Work Actually Looks Like

      An AI Product Engineer owns an AI feature from the user's problem to the production system: picking the model, writing the prompts and evals, building the product around it, and answering for its quality and cost after launch. Here's how the role differs from ML and software engineering, the skills it needs, and what the work looks like day to day.

      • AI Product Engineer
      • Career
      • AI Engineering
      • Product Engineering
      AI & Machine Learning7 min read
    • How to Run an LLM Cost Audit: Cut Your AI API Bill Without Hurting Quality

      AI spend usually grows from a few habits: one default model for every task, the same long prompt sent thousands of times, and nobody knowing which feature costs what. Here's the five-step audit I run on a codebase, with TypeScript examples for model routing, response caching and per-feature metering.

      • LLM
      • Cost Optimization
      • AI Engineering
      • Caching
      AI & Machine Learning8 min read

    Get the next Pulse by email.

    One email per edition, plus new articles.