Zapier Adds AI Model Switching to Prevent Vendor Lock-InZapier Adds AI Model Switching to Prevent Vendor Lock-InZapier Adds AI Model Switching to Prevent Vendor Lock-InZapier Adds AI Model Switching to Prevent Vendor Lock-In
August 14, 2026
Zapier has added model selection to every AI step in a workflow, letting a single automation route tasks through Claude, GPT, or Gemini across more than 400 connected AI tools without rebuilding anything when the provider changes. A third-party account places the change on May
Zapier has added model selection to every AI step in a workflow, letting a single automation route tasks through Claude, GPT, or Gemini across more than 400 connected AI tools without rebuilding anything when the provider changes. A third-party account places the change on May 20, 2026, though Zapier's own explainer reads less like a launch notice than an operating manual for avoiding vendor lock-in.
The dropdown menu is not really the story. The story is what it is built to prevent: vendor lock-in, a concern serious enough that a separate survey polled 542 U.S. C-level executives and decision-makers specifically on AI vendor dependency. For enterprise teams that have already let different departments standardize on different AI vendors, that is the difference between a policy memo and a technical guarantee.
What's new
Zapier, a workflow automation platform that connects thousands of business applications, has built model selection directly into AI by Zapier, its built-in tool for adding AI steps to a workflow (a "Zap"). Inside the tool, a dropdown menu lets a user pick which AI model runs a given step, or a given tool call inside an autonomous agent, and swap it for a different model in seconds without rebuilding the surrounding workflow. Users who need an action or configuration specific to one provider can instead use a "direct AI integration" for that model.
Zapier frames the result as flexibility rather than allegiance to any single AI maker: "With Zapier, you get flexibility. You can use whichever AI models you want, from whichever providers you want, and mix and match them in the same workflow depending on the task," the company said. The platform now connects more than 400 AI tools to its library of 9,000-plus everyday apps, and workflows can be mapped visually with Zapier Canvas, the company's diagramming tool for laying out multi-step, multi-model automations before they run. For teams wary of routing sensitive data through a vendor's own servers, Zapier also offers BYOM (Bring Your Own Model), which runs AI through a customer's own infrastructure, including Amazon Bedrock, while preserving existing security and compliance controls. IT teams get further guardrails on top of that: Managed Connections keeps app integrations under IT ownership as team membership changes, Domain Restrictions blocks personal accounts from connecting to business systems, and App Access Controls let admins allow or block specific apps workspace-wide. In one example workflow Zapier describes, a webinar transcript becomes a blog draft through Claude, then social posts through ChatGPT, in a single automated Zap.
Why it matters
The case for model flexibility rests on a simple observation: no single AI model wins every task, and teams that lock themselves into one provider inherit that provider's specific weaknesses along with its strengths. Zapier's own writers make the point in practice rather than theory. "I'm a Claude stan because it just gets my writing style, but I'll often reach for Sonnet over the higher-tier models because its results are more consistent for me," one Zapier blog author wrote, describing why a specific model, not just a specific vendor, ends up mattering for a specific job.
That logic maps onto how enterprise teams are starting to divide AI work by model rather than by department:
- Claude, specifically the Sonnet tier: writing tasks where consistency of voice matters more than raw capability
- Gemini: data-at-scale processing and multilingual triage
- GPT: classification and routing, plus general-purpose conversational tasks
Independent analyst commentary specifically on this announcement was not publicly available at publication time. What is measurable is how much of that model-routing work already runs through Zapier's own infrastructure: the company reports more than 3.39 million calls made through Model Context Protocol (MCP), the open standard, introduced by Anthropic, that lets AI systems talk to external tools and data sources. That is Zapier's own evidence that spreading AI work across models and tools, rather than betting on one, is already running in production rather than sitting in a roadmap.
Competitive Landscape
Zapier describes itself as "the most connected AI orchestration platform," a claim built on breadth rather than depth in any one model: more than 9,000 app integrations with governed access, an ecosystem that includes Google, Salesforce, and Microsoft, and what the company calls 13 years of production infrastructure behind the retry logic and error recovery that keep automations running when a single API call fails. The most direct competitive contrast the company draws is with OpenAI's Frontier, a product designed to build, deploy, and manage AI agents across an entire business from within a single provider's stack, the deeper-lock-in model Zapier is positioning against.
Independent rankings of Zapier's broader competitive set, the low-code automation tools that compete on price or code access rather than on cross-model orchestration, were not available in verified source material for this piece. What is confirmed is the shape of the contrast Zapier itself draws: a single-vendor, full-stack agent platform like Frontier against a many-vendor orchestration layer that treats the model itself as a swappable component. That distinction, not pricing or app count, is the axis Zapier is competing on.
What's next
The scale numbers behind Zapier's AI push are still climbing. The company's live counter puts cumulative AI tasks automated on the platform at 593,138,971 and counting, tracked since January 2023. Zapier also reports more than 450,000 agents built on the platform and more than 3.39 million MCP tool calls completed, both signs that the model-switching feature is landing on top of an already-active automation base rather than seeding a new one.
The next layer is aimed at builders rather than end users. Log Streaming sends real-time workflow data to external monitoring tools, named as Datadog and Splunk, so engineering teams can watch AI-driven automations the same way they watch any other production system. On the developer side, Zapier connects coding agents and AI assistants through two paths it describes as "different doors, same building": the Model Context Protocol for AI assistants, and a software development kit for coding tools including Claude Code and Cursor, both routed through one shared authentication layer and a single admin log. What Zapier has not published is pricing detail for AI by Zapier against a direct provider integration, or whether the model-switching dropdown is available across all workspace tiers or gated to specific plans.
The switching cost that vendor lock-in research keeps returning to is rarely the sticker price of a new tool; it is the sunk cost of everything already wired around the old one. Zapier's bet is that the dropdown menu, not the model behind it, is the asset worth owning.
For a CIO evaluating AI spend across a dozen departments, the practical shift is procurement, not features: instead of negotiating separate contracts, security reviews, and SSO (single sign-on) configurations for every team that wants to standardize on a different model, one set of controls, SOC 2 (Type II) and SOC 3 certification (independent audits of a vendor's security practices), SAML-based single sign-on, and SCIM (automated user account provisioning), covers all of them at once, alongside GDPR and CCPA (EU and California data-privacy law) compliance already built in. A marketing team running Claude for copy and a data team running Gemini for multilingual triage stop being two vendor relationships and become two dropdown selections inside one already-audited platform. Whether that convenience holds once OpenAI, Google, or Anthropic decide orchestration itself is worth competing on directly is the open question 593 million completed tasks have not yet answered.
-- Aria Lin, Enterprise Technology Analyst
Sources: Zapier · Model Context Protocol · System and Organization Controls (SOC)