Skip to content
Agentdex

Model Pulse #4 · Tuesday, Oct 27, 2026

Today in AI: agents become code, and your Downloads folder becomes an attack surface

The seven AI stories from the last two days that matter for AI coders and indie game devs. Each one links to its primary source.

Sponsored Sponsor this issue: reach AI coders and solo founders in Model Pulse.

Oct 26, 2026

NEW: Cursor's Agent SDK lets you define agents as code

Cursor made its Agent SDK generally available on Oct 26. An agent is a folder: prompts in Markdown, tools in TypeScript. You can deploy it on Cursor's hosted infrastructure or run it yourself, and built-in evals measure every change. Cursor says the same setup powers its own agents such as Bugbot. For AI coders, this means custom review, triage or build bots can live in your repo and go through code review like everything else, instead of being configured in a dashboard.

Source: https://x.com/cursor_ai/status/2114793714531102901

Oct 22, 2026 (reported Oct 25)

SECURITY: A planted AGENTS.md in your Downloads folder can hijack coding agents

Forever Security reports it caught a malicious AGENTS.md in a customer's ~/Downloads folder that told OpenAI's Codex to add an MCP server to its machine-wide ~/.codex/config.toml and install the known-malicious npm package function-color. Forever blocked it. Replaying the real file in an isolated VM, Codex ran the malware in more than 90% of attempts, and a similar trick could leak Claude Code's login token. Websites can drop files into Downloads without a click, so the fix is on your side: never start an agent in Downloads, read any instruction file you didn't write, and check your agent config for servers you didn't add. The test numbers come from Forever and have not been checked independently. BleepingComputer covered it on Oct 25.

Source: https://forever.security/blog/hijacking-ai-agents-via-downloads-folder/

Oct 26, 2026

BIG MOVE: Anthropic files for its IPO

Anthropic filed a Form S-1 with the SEC on Oct 26 and plans to list Class A shares on the Nasdaq under the ticker "ANT". The share count and price range are not set yet. Fortune reports the prospectus shows $16.3 billion in revenue for the first half of 2026 and an $80.7 billion net loss, mostly from non-cash accounting charges, with an operating loss of $693 million. If you build on Claude, a public Anthropic will have to report numbers every quarter, which should make its pricing and capacity plans easier to read.

Source: https://www.anthropic.com/news/s1-sec

Oct 25, 2026

OPEN WEIGHTS: Nanbeige4.3-33B targets local coding agents

Nanbeige LLM Lab, the AI team at Chinese job platform BOSS Zhipin, released Nanbeige4.3-33B under Apache 2.0, with a 512K context. It was announced on Oct 25. The lab's own numbers put it at 80.9 on Terminal-Bench 2.1 against 73.0 for Qwen3.8-27B, and 65.5 against 42.5 on DeepSWE. Qwen still wins slightly on SWE-Bench Pro. These are self-reported scores, not yet independently checked. A smaller distilled version is planned. If you run agents locally, it is worth testing against your current ~30B model.

Source: https://huggingface.co/Nanbeige/Nanbeige4.3-33B

Oct 26, 2026

CHEAPER: DeepSeek V4.1 Flash lands in Cursor

Cursor added DeepSeek V4.1 Flash on Oct 26. Cursor says every effort level sits on its CursorBench cost frontier, at $0.60 to $0.98 per task, with no model scoring higher at those prices. You turn it on in Cursor Settings > Models. It is a cheap option for routine edits, so you can save frontier models for the hard problems. The benchmark is Cursor's own.

Source: https://x.com/cursor_ai/status/2114839997568405608

Oct 26, 2026

GAME DEV: Zero Escape creator Uchikoshi says not using AI can cost jobs too

In a Famitsu interview published Oct 26, Kotaro Uchikoshi (Zero Escape, AI: The Somnium Files) explained why his new three-person studio CRASH KEYS went public early: a game his team spent about 18 months on was canceled. He put it down to a change of plan at the publisher, in a tough market where solo indie devs using AI now compete with mid-sized adventure-game teams. He backs using AI for repetitive work, like endless small visual tweaks, while people keep the story and design decisions, to cut development time rather than staff. Some fans have pushed back. For indie devs, the takeaway is that the AI argument has reached established studios too.

Source: https://www.famitsu.com/article/202610/92056

Oct 26, 2026

DEALS: $63 billion to put Broadcom chips to work for Anthropic, and Nevis hits a $500M valuation

Bank of America, Blackstone, Citi and Morgan Stanley announced a $63 billion financing on Oct 26. A new company will buy Broadcom AI accelerators and networking gear and lease them to Anthropic, with deliveries from 2027 and Broadcom backing part of the risk. Reuters called it one of the largest debt deals of the AI buildout. Also on Oct 26, Nevis, whose AI agents handle CRM, meetings and planning work for financial advisors, raised a $70 million Series B led by EQT Ventures. That brings its total to $110 million, and Fortune reports a $500 million valuation. Investors are still writing big checks for compute and for AI agents in specific industries.

Source: https://investors.broadcom.com/news-releases/news-release-details/bank-america-blackstone-citi-and-morgan-stanley-announce

Ship faster this week

Grab the free Agentdex starter kit with prompts and agent configs for Cursor, Claude Code and Codex.

Get the free kit

Model Pulse

Daily AI model drops, tools, and what they mean for AI coders.

We store your email with Resend and use it only for Model Pulse. See privacy. Missed the last issue? Read Model Pulse #3. Next issue: Model Pulse #5.