Anthropic proposes a growing public-benefit stake

Anthropic proposes putting 5% of its equity toward public benefit

Anthropic announced a Public Benefit Fund on October 18, with an initial commitment of 5% of its shares subject to stockholder approval. It is a substantial proposal for distributing AI’s economic gains, not a completed payout to the public.

• For every 50% increase in company value above a set baseline, the board could add another 2% of shares at its discretion.
• The fund will be a wholly owned LLC. Possible uses include nonprofit donations or distributions to specific populations; its equity cannot fund lobbying or political giving.
• The roughly 20% headline figure includes existing founder and employee pledges. It is not the new fund’s starting stake.
• Governance details and valuation milestones are still forthcoming, so who receives support and how decisions get made remain open questions.

What it means The proposal puts AI’s economic benefits into a concrete equity structure, but approval and governance will determine whether that structure delivers meaningful public support.

Source: www.anthropic.com

Beam’s weights and coding harness are now open

Reflection released Beam’s weights under Apache 2.0 on October 16, alongside Mirror CLI, the open-source agent harness it trained with. You can now self-host and customize the model rather than depend solely on hosted access.

• Beam has 501 billion total parameters, with 23 billion active. The release includes BF16, FP8 and NVFP4 weights plus a speculative-decoding drafter model.
• Launch support includes Transformers, vLLM and SGLang, giving teams existing serving paths rather than requiring custom inference plumbing.
• Mirror CLI lets you inspect agent activity, approve tool calls and steer work. It also supports resumable conversations and parallel agents in separate worktrees.

What it means You can modify both the model and its trained agent harness, giving self-hosted coding systems control over deployment, tool approvals and workflow behavior.

Source: reflection.ai

DeepSeek opens its Search and Fetch APIs to everyone

DeepSeek opened public access to its in-house Search and Fetch APIs on October 17. Outside builders can now use infrastructure that the company says already serves hundreds of millions of requests daily.

• Each endpoint costs $0.75 per 1,000 calls. Search and Fetch are priced separately.
• For DeepSeek API users, calls through server-side tools in DeepSeek Harness, Claude Code and Codex are currently free.
• That free period ends when billing starts on October 31, 2026, at 16:00 UTC.

What it means You have another public search-and-fetch backend for agents, with explicit per-call pricing; budget both endpoints separately rather than treating the temporary harness exemption as permanent.

Source: api-docs.deepseek.com

AI workflow of the day

How to draft meeting follow-ups in your own voice

Shared by Ruben Hassid (@rubenhassid) · View original post ↗

For operators juggling meetings and email, this Claude workflow turns meeting notes into a follow-up draft using examples of how you actually write. You keep control of what gets sent.

1. Connect Granola and Gmail to Claude so it can access your meeting notes and sent emails.
2. Ask Claude: "Use Granola connector, go to my last meeting with [name]. List down the deadlines, tasks, and blockers. But clearly state if something wasn't said clearly. Do not guess."
3. Build your writing profile: "Read my last 30 sent emails in Gmail. Write my voice profile: how I greet, how I sign off, my sentence length, my go-to phrases, the words I never use. Format it as writing instructions to follow when drafting as me."
4. Use the prompt below to request the follow-up as a Gmail draft.
5. Review the draft yourself before sending. Do not let Claude send replies automatically.

Pull my last meeting from Granola. Draft the follow-up using my voice profile: what we decided, action items with owners, one clear next step. Keep it as short as I keep mine. Create it as a Gmail draft to the attendees.