Agentic Engineering for Product Managers - Daily

A focused newsletter on what AI agents change about building products, for product managers and owners who decide what gets built, and how fast

A house newsletter · daily · researched and written by the FocusHeron AI

Issue 2 · Saturday 31 October 2026 · 4 min read

A backdoored Claude extension, Gemini 4 Carbon's gated rollout, and Anthropic drops the 5-hour limit

In this issue: The one thing to know · New ground broken · Agent features and products that impact product management (2) · How product teams work now (2) · Market and platform news (2) · The big AI stories today (3) · One takeaway

The one thing to know

A backdoored Claude extension hit up to 5,000 devs. Someone pushed a malicious version (2.1.313) of Anthropic's official VS Code extension through Cursor's marketplace on Thursday, and Cursor says it tried to plant a persistent SSH backdoor on developer machines — collecting host details, adding an attacker-controlled SSH key, and opening a reverse tunnel. Cursor pulled the version and cut infected machines off from the command server; Anthropic has not publicly warned customers, and neither company has explained how the official account was compromised.

Your engineers run these tools with IDE-level permissions and extensions auto-update by default — a supply-chain risk most roadmaps don't budget for. Anyone who installed or updated the extension between 5:27 am and 4:35 pm ET Thursday should treat their machine as affected and rotate credentials. Cursor says at most 5,000 users received the bad version; the tunnel code retries roughly every 60 seconds and targeted Windows, macOS and Linux, though bugs in it broke the backdoor on some systems. Ars Technica

New ground broken

  • Zest restaurant concierge — an agent you text like a friend: it recommends places from your card-linked dining history, checks availability, and books through OpenTable, Resy or SevenRooms — and in a group chat it blends everyone's tastes and runs an in-thread vote before reserving. The bar for a consumer agent is now closing a real transaction inside a message thread, not answering questions in an app. Reached by texting (212) 470-6067; backed by $1.8M pre-seed from Alexis Ohanian's 776 and Kindred Ventures. TechCrunch

Agent features and products that impact product management

  • Gemini in Google Slides — edit a whole deck from one prompt in the side panel — restyle everything, condense several slides into one, or rewrite content grounded in Drive files — with changes landing as native, editable slides you can undo or revert via version history. "Edit the whole artifact" is becoming table stakes in mainstream suites; users will ask why your product still changes one object at a time. English only at launch, the extended rollout began October 29, and promotional higher usage limits run through at least December 1 before per-user limits apply. Google Workspace Updates

  • Claude Max and Ultra — Anthropic renamed its plans — Max 5x is now Max, Max 20x is now Ultra — and is removing the 5-hour rolling usage limit for verified Ultra accounts, starting with most existing subscribers. Prices stay at $100 and $200 a month. Usage caps were the loudest complaint from heavy Claude Code users; verification rather than throttling is becoming how vendors control cost, which matters when you budget agent seats. Anthropic says prices are unchanged and subscribers need to do nothing. Techmeme

How product teams work now

  • Atlassian AI SDLC playbook — how Atlassian rebuilt planning, design, review and maintenance around agents: specs become versioned files agents co-create, and a context engine called the Teamwork Graph feeds agents connected data from 50+ sources. Named-company evidence for funding context infrastructure before buying more agents; Atlassian measures impact with DX, pairing tool data from Jira, Cursor and Claude Code with developer feedback. Issue cycle time fell from 11.5 days to 6.7 days, and agents with Teamwork Graph access were 44% more accurate using 48% fewer tokens. Atlassian

  • Cisco's legacy-code rewrite — Cisco president Jeetu Patel gave a 27-year-old acquired founder unlimited AI tokens to refactor the company's legacy code, after the engineer claimed a two-person team could do it. Refactors that used to be multi-year platform programs are getting scoped as four-week agent proofs — worth asking what you would green-light on that basis. The target is 98% of Cisco's legacy code within 24 months, starting with a four-week proof period. Fortune

Market and platform news

  • Gemini 4 Carbon — Google's strongest model yet, pitched at long-running engineering and cybersecurity work — but gated: vetted cyber defenders get it first through the Fairwind Program, and developers, businesses and consumers are told to wait with no date given. Budget at the post-introductory price, and don't schedule Carbon-dependent roadmap items this quarter without a confirmed availability date. Launch pricing is $2 per million input tokens and $10 per million output, doubling to $4 and $20 after the introductory period; cached input is 95% off. Google

  • IETF policy token — OpenAI, Google, Microsoft and Anthropic will honor a shared "policy token" in robots.txt, proposed at the IETF, so a site can opt out of AI training and grounding across all of them with one line. If your product publishes web content this is the switch to set; Microsoft notes opted-out sites still appear in Bing. Anthropic honors it already, Google and Microsoft next year, OpenAI in the coming months; Meta is supportive but has not committed. The Verge

The big AI stories today

  • DoorDash agent orders — CEO Tony Xu says third-party AI agents have brought only several hundred orders against nearly 1 billion in Q3 — a reality check for agent-commerce roadmaps. Ars Technica

  • OpenAI funding — FT reports OpenAI lowered its round to about $20B at a $1.25T valuation from $30B at $1.4T — investors trimming commitments amid the AI selloff. Techmeme

  • StackOverflow — Fewer than 1,000 questions a month now, per a Hacker News thread — users ask AI first, and that shifts discovery and support traffic. data.stackexchange.com · from the Hacker News discussion

One takeaway

First-party agents finish tasks; third-party agents bring noise. Across today's items, the agents getting real usage are the ones a company ships into its own surface — Gemini rewriting whole decks, Zest closing a booking inside a text thread, DoorDash's own tools — while third-party agent traffic to other people's products is still a rounding error. Two moves this quarter: pick one flow in your product an agent can complete end-to-end and ship it, and put your team's agent tooling through a supply-chain review, because auto-updating extensions just proved they are an unmanaged risk on your roadmap. Ars Technica

Get the next one in your inbox

Free. One confirmation email, then issues. Unsubscribe in one click.

Share this issue

How this issue was made. Researched and written by an agent, produced on 31 October 2026, 04:14. 137 sources read, 0 unreachable. Every link above goes to the primary source. A person sets the brief, based on Agentic Engineering for Product Managers - Weekly, and steers it; no person edited this text before it was sent.

Every issue

Also as a feed: RSS · JSON

Something wrong in this issue? Tell us