- Monday Momentum
- Posts
- The Convergence Wars
The Convergence Wars
Why OpenAI's AgentKit launch won't kill Zapier, and how the industry transformation is good for everyone
Justin Wright
October 13, 2025 • Est. Reading Time: 12 minutes
Happy Monday!
Last week, I explored how AI-generated "workslop" reveals that successful AI adoption requires organizational transformation, not just tool deployment. But while companies struggle with implementation, a shift in workflow automation is accelerating: the evolution from connecting apps to orchestrating autonomous agents.
OpenAI launched AgentKit at DevDay, a complete toolkit for building and deploying AI agents with visual workflow design, connector management, and evaluation tools. An OpenAI engineer built a working agent in under eight minutes on stage. Ramp reported going from "months of complex orchestration" to "just a couple of hours" with Agent Builder.
The initial reaction: "OpenAI just built a Zapier competitor." The reality is more nuanced, and truthfully more interesting. AgentKit represents the third generation of workflow automation, moving from trigger-action pairs to autonomous decision-making. This doesn't kill existing platforms; it forces them to evolve while serving fundamentally different use cases. The question isn't who wins, but how the entire category transforms.
OpenAI's AgentKit offers visual agent building, connector management, and evaluation tools, creating agents in minutes versus hours with traditional platforms. Rather than killing Zapier's 8,000-app ecosystem or n8n's code-first flexibility, AgentKit validates the agentic workflow category while serving different needs: model-native development versus broad integration orchestration.
The Three Generations of Workflow Automation
Understanding AgentKit's impact requires seeing how workflow automation evolved through distinct technological eras, each solving different problems with different approaches.
Generation 1: Integration-First (2011-2020)
Zapier launched in 2011 with a simple premise: connect apps without coding. The model was based on trigger-action pairs: when something happens in App A, do something in App B. Zapier now connects nearly 8,000 apps, making it the dominant iPaaS (integration platform as a service).
The value proposition was clear: eliminate manual data entry, sync information across systems, and automate repetitive tasks. Companies like Contractor Appointments reported helping clients generate $134 million in revenue through automation.
n8n emerged as the developer-friendly alternative, offering open-source workflow automation with code flexibility. While Zapier prioritized ease of use for business users, n8n gave technical teams full control with JavaScript/Python support and self-hosted deployment options.
The limitation: these tools moved data efficiently but couldn't make decisions autonomously. Every workflow required explicit human logic i.e. “if this happens, then do that.” Complex workflows became brittle as edge cases multiplied.
Generation 2: AI-Enhanced Automation (2020-2024)
The arrival of capable LLMs transformed what workflows could do. Zapier introduced AI steps, allowing workflows to generate text, analyze sentiment, or summarize content. n8n added native AI nodes for LangChain integration and custom model connections.
Zapier launched Agents in 2024 which were tools that could "improvise a little" rather than following rigid logic. The difference: an agent might research a lead, determine the best approach, and craft personalized outreach. These types of tasks require judgment, not just data movement.
The challenge: these AI enhancements sat on top of fundamentally integration-focused architectures. As Zapier's own documentation notes, "Choose Zaps when you need precision and predictability. Choose agents when 80% accuracy is genuinely sufficient."
Generation 3: Agentic Orchestration (2024-2025)
AgentKit represents a key architectural shift: building for autonomous agents first, integrations second. The platform includes Agent Builder for visual workflow design, Connector Registry for tool management, ChatKit for embeddable UIs, and critically, comprehensive evaluation tools including trace grading, automated prompt optimization, and third-party model support.
The architectural difference matters. Traditional platforms ask: "How do we add AI to workflow automation?" AgentKit asks: "How do we orchestrate multiple AI agents with the reliability of traditional workflows?"
Pattern Recognition: Convergence from Multiple Directions
The workflow automation market is experiencing convergence from three distinct vectors, each bringing different strengths and limitations.
Pattern #1: Model Providers Building Orchestration
OpenAI's move into workflow automation follows Amazon's AWS playbook: solve internal problems, package the solution, sell it as a platform. Klarna built a support agent handling two-thirds of all tickets. Clay achieved 10x growth with a sales agent. These successes required internal orchestration tools that ultimately became AgentKit.
The competitive moves are coordinated. Google launched its Agent Development Kit in April, enabling multi-agent system building "in under 100 lines of code." Microsoft also announced the Agent Framework, bringing agent creation into one place. Model providers recognize that controlling the orchestration layer matters as much as providing models.
The advantage: native model access, built-in evaluation, and deep integration with the provider's AI ecosystem. The limitation: narrower integration ecosystems and platform lock-in concerns.
Pattern #2: Integration Platforms Adding Intelligence
Zapier rebranded as an "AI orchestration platform," explicitly positioning itself beyond simple integration. The company describes AI orchestration as "the connected, end-to-end application of AI tools, agents, and automations across workflows, teams, and systems."
Zapier's latest features include AI by Zapier with model choice, Agents in Zaps for research-heavy tasks, and Global Variables for scaling automations. The positioning is clear: Zapier argues that traditional workflow orchestration creates systems where tasks execute in order, while AI orchestration adds intelligent decision-making to workflows.
n8n evolved similarly, positioning as a platform that "uniquely combines AI capabilities with business process automation". The company emphasizes its strength: "The flexibility of code with the speed of no-code" for technical teams. Recent pricing changes removed workflow limits, encouraging experimentation with complex AI systems.
The advantage: massive existing integration libraries, established customer bases, and proven reliability at scale. The limitation: AI capabilities layered onto integration-first architectures.
Pattern #3: Developer Tools Becoming Platforms
The third vector comes from developer frameworks like LangChain and AutoGen becoming more user-friendly. While these tools remain code-first, they're adding visual builders and no-code elements that blur the line between framework and platform.
This convergence creates interesting competitive dynamics. Everyone is racing to own the complete stack, but from different starting points with different core competencies.
The False "Zapier Killer" Narrative
The immediate reaction to AgentKit was predictable: "OpenAI just became a Zapier competitor." Some developers noted that AgentKit could compete with platforms like Zapier. But several developers viewed AgentKit not as a Zapier killer, but as a tool that complements the pipeline.
The comparison misses fundamental differences in use cases, audiences, and architectural priorities:
Integration Breadth vs. Model Depth
Zapier's 8,000 integrations represent 14 years of partnership development, API maintenance, and edge case handling. Each integration requires ongoing support as APIs change, apps update, and user needs evolve. This ecosystem is Zapier's moat.
AgentKit offers a Connector Registry for managing tool connections, but with far fewer pre-built integrations. The trade-off: deeper native integration with OpenAI's models, comprehensive evaluation tools, and optimization specifically for agentic workflows.
For a sales team automating lead enrichment across HubSpot, Clearbit, LinkedIn, and internal databases, Zapier's breadth wins. For an AI research team building multi-agent systems with complex reasoning chains, AgentKit's model-native architecture wins.
Business Users vs. AI Developers
Zapier's positioning emphasizes that "anyone at your organization" can leverage AI, regardless of technical expertise. The platform hides complexity, offering pre-built templates and wizards that guide non-technical users through automation creation.
AgentKit targets developers and technical teams comfortable with concepts like trace grading, prompt optimization, and multi-agent workflows. The platform includes ChatKit for embedding agents, but assumes developers will build the wrapper applications.
n8n occupies middle ground, appealing to technical teams who want code-level control without building everything from scratch. The platform's open-source nature means developers can modify, extend, and self-host, addressing data sovereignty concerns that many enterprises have with cloud-only solutions.
Deterministic Reliability vs. Agentic Flexibility
Zapier's core strength remains deterministic workflows: when X happens, always do Y. This predictability matters for business-critical processes. Finance teams don't want agents that "improvise a little" when processing invoices.
AgentKit embraces the non-deterministic nature of AI agents, providing evaluation tools to measure and improve performance rather than guaranteeing identical outputs. Evaluation features include datasets, trace grading, and automated prompt optimization, all acknowledging that agentic systems require different quality assurance approaches.
Contrarian Take: AgentKit Validates the Category, Doesn't Own It
Rather than threatening Zapier and n8n, AgentKit's launch validates their strategic evolution while accelerating the entire category's maturation. History shows that infrastructure providers entering application layers rarely kill established platforms; they instead force evolution.
Consider analogous market dynamics:
AWS vs. Salesforce: Amazon entered the application layer with numerous SaaS offerings, yet Salesforce not only survived but thrived by building on AWS while maintaining platform dominance. Why? The "last mile" of enterprise adoption, customization, integration, change management, remains hard even with great infrastructure.
Twilio vs. WhatsApp Business: Facebook (Meta) offering business messaging didn't kill Twilio; it expanded the communications market while serving different needs. Developers choosing between platforms consider factors beyond raw capability: ecosystem, flexibility, and control.
Stripe vs. Bank APIs: Banks offering APIs didn't eliminate payment processors; the abstraction layer, developer experience, and workflow integration justify the middleware margin.
AgentKit faces similar dynamics. Zapier's large number of integrations, established customer base, and workflow templates represent switching costs that take years to replicate. n8n's open-source community, self-hosted deployment options, and code flexibility serve needs that closed platforms cannot.
The real impact: AgentKit forces faster evolution. Zapier must accelerate its AI orchestration capabilities. n8n needs to strengthen its agentic tooling. Both must articulate why their approaches matter in an agentic world. This competition largely benefits users.
Orchestration Becomes Table Stakes
The convergence reveals where the market is heading: workflow automation as a discrete category is disappearing, replaced by agentic orchestration as a fundamental capability embedded across the stack.
The question facing every platform:
For model providers: Can you build sufficient integration breadth and developer tooling to compete with established workflow platforms?
For integration platforms: Can you evolve from moving data to coordinating intelligent agents while maintaining reliability?
For open-source tools: Can you match the evaluation, monitoring, and governance capabilities that enterprises need for agentic systems?
The winners will be platforms that recognize these tools serve different audiences with different needs. Zapier's positioning around AI orchestration acknowledges this shift. n8n's emphasis on mixing "AI, code, and human steps in a reliable way" reflects understanding that agentic systems need structure. AgentKit's comprehensive evaluation tools address the governance gap that prevents enterprise adoption.
The broader implication: the "orchestration layer" becomes as critical as the model layer. Just as database selection (MySQL, PostgreSQL, MongoDB) depends on use case rather than one being universally superior, orchestration platform choice will depend on whether you prioritize integration breadth, code control, or model-native tooling.
For developers and enterprises, this means thoughtful platform selection based on actual needs rather than hype. For investors, it suggests a larger market with room for multiple winners serving distinct segments. For the AI industry, it validates that infrastructure and tooling matter as much as model capabilities.
In motion,
Justin Wright
If OpenAI can build agents in 8 minutes that previously took days with traditional platforms, but Zapier maintains an 8,000-app ecosystem that takes years to replicate, does this suggest the real competitive moat in workflow automation is actually breadth of integration and the organizational switching costs that come from embedding workflows across teams?
Sora update #1 (Sam Altman)
A collaborative approach to image generation (Google)
OpenAI valuation reaches $500 billion (Bloomberg)
A Command Line Companion for Google’s Coding Agent (Google)
AMD and OpenAI announce strategic partnership (OpenAI)
An open-source auditing tool to accelerate AI safety research (Anthropic)
Full OpenAI Dev Day video (YouTube)
I am excited to officially announce the launch of my podcast Mostly Humans: An AI and business podcast for everyone!
Episodes can be found below - please like, subscribe, and comment!