Merge helps companies build connected software and AI products, from syncing customer data to enabling AI agents to take action across business systems. Its products span customer-data integrations through Merge Unified, secure tool access for AI agents through Merge Agent Handler, and model routing and cost controls through Merge Gateway. Merge for Workforce brings those capabilities inside companies, helping teams manage which models, tools, and business systems employees can access. Together, these products support both the software companies build for customers and the AI-powered work happening inside their organizations.
Inside Merge, CEO and Co-founder Shensi Ding uses Codex to review product code, including work on a new setup interface for configuring tool access in Merge for Workforce. Her colleagues use GPT-6 Astra for product testing and marketing. The team also uses AI for the work of running the company, from preparing sales updates to planning customer dinners. "We use it honestly for everything. Integrations obviously, but we also use it to automate different workflows, to help make our product better overall, to incorporate any feedback from our customers, and to build end-to-end features." — Shensi Ding, CEO and Co-founder, Merge
Inside the rollout, Merge launched a tool for adding connectors to AI products, with reliability a priority for Ding: "We’re pretty proud of the quality. Tool calls through Merge succeed 99.9% of the time, compared to public MCP servers, which are around 60%." For its own product development, Merge uses GPT-6 Astra. "It’s really fast, and it was much better at code reviews," Ding says. "They also liked the personality of GPT-6 Astra, which goes a long way. It was also clear that Codex had improved a lot. Merge lets employees choose their own AI tools, and which one to use has become a lively internal debate."
Ding used Codex to review a proposed implementation of the new provisioning tab in Merge for Workforce as the team evaluated how customers would configure tool access. They discussed making the "Add tool pack" action a clearer link that opens separately to make setup less confusing. Codex supplied code-review feedback; Merge’s team made the product decisions and implementation changes, with engineers responsible for validating and shipping them. "Other AI systems would end up forgetting what I told them and only doing the last thing that I said. GPT-6 Astra continues to execute on every single task really well, and it fills in the gaps for what I did not explicitly say using common sense." — Shensi Ding, CEO and Co-founder, Merge
The dinner bot Merge hosts executive dinners for customers and prospects to build relationships and create opportunities for follow-up product conversations. For one dinner in Dallas, the bot pulled potential guests from the CRM and Apollo, checked previous invitations and responses, and helped prepare outreach from Ding’s email. It also maintained the attendee tracker, proposed a seating chart, and created calendar notes where sales and customer success could add context about each guest. The model handles reasoning and drafting, while Merge’s connectors provide access to the business systems. Ding reviews the output, rewrites anything that sounds off, and curates the guest list using relationship context the bot does not have. In October 2026, Merge reported that preparing guest lists and outreach and maintaining the attendee tracker had gone from dozens of hours across the company to about 30 minutes per event. The comparison covers the same planning tasks, measured over the preceding three months. The team still handles hosting, relationship-building, and on-the-ground logistics. Merge ran 16 executive dinners in Q3, up from 8 in Q2, and plans to host more than 30 in Q4. After the Dallas dinner, the team reported positive feedback on the invitations and opportunities for follow-up product conversations.
Source: OpenAI Blog · Summarized by HeadlinesBriefing