Every company manages two layers of AI. One helps employees do their work and the other runs inside the products customers use. Both need to be secure and connected to the right data. This fall, Merge launched a product for each: Merge for Workforce and Merge for Products. In this recap @aj2w and @sarahraemazur, who lead product and engineering, demo each one end to end. Merge for Workforce securely brings approved skills and company context into the coding agents employees already use. IT sets model budgets and plain-English guardrails once, like blocking confidential customer details from an agent’s Slack message. Merge for Products gives teams building AI features Gateway for models, Agent Handler for tools, and Unified for customer data, with one setup every new customer reuses. What have we been cooking:

Ask an agent "which customers were upset last week?" and a few MCP servers won't get you there. Live API lookups can't search across that much data. At the AI Engineer World's Fair, Gil Feig (@GilFeig) walked through how to build a context graph that syncs data for semantic search and keeps live calls for simple lookups.

Merge Agent Handler just got 11 new connectors! - Salesforce Marketing Cloud (@salesforce) - Salesforce Data 360 (@salesforce) - Adobe Experience Manager (@Adobe) - Adobe Workfront (@Adobe) - Excel GCC High (@Microsoft), for US government tenants - Tavily (@tavilyai) - Kit (@kit) - Checkly (@checklyhq) - ConfigCat (@configcat) - Alpha Vantage - TinyPNG (@Tinypng) Try them out:

Mistral Large 4 from @MistralAI is now available on Merge Gateway. It's a big jump for Mistral. Large 4 scores 38 on the Artificial Analysis Intelligence Index, more than double their previous best. Open weights ship this month.

The biggest generational jump since June came from a model that costs $0.18 per million output tokens. DeepSeek-V4-Flash-0731 (@Deepseek_AI) scored 26.8 points higher on Toolathlon than the release it replaced. GLM-5.3 gained 24.8 on the same board. Claude Opus 5.5 added 20.7 on SWE-bench Pro, and Claude Sonnet 5.5 added 18.1 there six days later. Scores via week of Sep 28.

We're looking for an AI pilled PM to join the team and work directly with me on owning our newest initiatives. Apply if you're perpetually on X, curious about the latest tools, and want to build without barriers for some of the world's largest enterprises and AI companies.

Enterprise AI shouldn't force teams to choose between control and flexibility. Merge Gateway now supports self-hosted models so that teams have a single place to operate them alongside models from @OpenAI, @AnthropicAI, and countless other providers. Keep sensitive workloads on your own infrastructure, while governing everything through Gateway. Check it out:

GPT-6.1 Sol from @OpenAI is now available on Merge Gateway. Astra level coding at a fifth of the price. It ties GPT-6 Astra on DeepSWE v1.1 and scores 6.4 points above GPT-6 Sol. Check it out:

The model that tops agentic tool calling costs $0.50 per million output tokens. GLM-5.3-Flash leads Toolathlon at 78.4%, and five of the top six run under $5 per million out. Price and tool-calling accuracy have come apart at the top of this board. That's an argument for routing on the benchmark instead of committing to one model up front. Weight Toolathlon in a Build-Your-Own-Router policy and Gateway sends each tool-heavy request to whatever leads it that week. Scores via