|
|
|
Claude adds 43 free workflows, Siri goes Gemini, and a 744B open model lands MIT
|
|
Cline shipped a standalone Desktop app, made Hub-managed Agent Plugins general so MCP servers auto-discover from ~/.agents/plugins instead of hand-edited JSON, and added turn retry that survives a forwarded 429. ClinePass did not move off $9.99.
|
|
|
|
|
Shanghai AI Lab puts a 744B agentic model under MIT
Atria Dawn Preview is a mixture-of-experts agent built on GLM-5.2, weights on HuggingFace. The preprint claims top scores on five of 16 benchmarks; in a study of 769 task records across 56 participants, roughly one in three completed AI-assisted tasks were rated infeasible without AI. The license is the part that changes your options: this runs where an outbound API call is not allowed. Weights on HuggingFace →
|
Anthropic expands Claude for Small Business to 43 workflows and 27 integrations
Installs are past 900,000 since May. The 27 added connectors include Zapier, Shopify, Salesforce, Xero, Gusto, Square, and Stripe. Every workflow still opens in approval mode - Claude drafts and stages the work, then stops until the owner sends, posts, or pays. Free on any paid Claude plan; permissions inherit from the connected app. Full breakdown →
|
GitHub Copilot lets you set the cost-quality tradeoff, not the bill
Auto model selection gained three tiers - efficiency, balance, and intelligence - in VS Code, Copilot CLI, and the Copilot app. You are charged for whichever model auto picks regardless of tier, so efficiency saves money only insofar as it routes you somewhere cheaper. Set balance, then check your usage dashboard a week later. Try Copilot →
|
OpenAI, Anthropic, and Google admit weeks of safety coordination
OpenAI policy chief Chris Lehane said September 15 that the three firms have been in safety talks for weeks, and they do not need the antitrust waiver Dario Amodei proposed. King Charles convenes roughly 30 AI executives at Dumfries House this week with a Ditchley Foundation charter on the table. Voluntary standards get drafted before statutory ones. Summit preview →
|
|
•
|
Make. Claude's small-business pack speaks Zapier now, but Make fills the last mile for anything Claude does not connect to - free tier covers one end-to-end run. Try Make →
|
|
•
|
CustomGPT. Point it at your docs and it answers from your content instead of the open web - quickest way to test support-bot deflection before committing to a platform. Try CustomGPT →
|
Cline is a better buy in September at the same $9.99
Cline shipped a standalone Desktop app, made Hub-managed Agent Plugins general so MCP servers auto-discover from ~/.agents/plugins instead of hand-edited JSON, and added turn retry that survives a forwarded 429. ClinePass did not move off $9.99. Solo dev: run free with your own key, add ClinePass once rate limits cost you more than $10 a month. Team of 2 to 10: $9.99 a seat undercuts $20 Cursor Pro, provided open-weight quality clears your bar on tangled refactors. Read the September buy call →
|
|
|
|
|
Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.
|
|
|
Affiliate disclosure
·
Pondero earns commissions on some links. This does not affect our editorial picks.
|
|