Presenters
Source
Zapier’s AI Revolution: Orchestrating the Future of Work 🤖✨
Welcome back to the blog, tech enthusiasts! Today, we’re diving deep into the fascinating world of AI-powered automation with Chris Geoghegan, VP of Product at Zapier. Zapier, a company that needs no introduction in the automation space, has become the backbone for over 3.4 million businesses, connecting over 7,000 applications to streamline workflows without a single line of code. With a $5 billion valuation and impressive profitability since 2014, Zapier is not just keeping pace with the AI revolution; they’re actively shaping it.
In this insightful conversation with Carlos Gonzalez de Villaumbrosia, CEO at Product School, Chris shares how Zapier is leveraging AI, building “agentic workflows,” and navigating the complexities of enterprise adoption. Let’s unpack the key takeaways!
From Triggers to Workflows: Zapier’s Evolution 🚀
Zapier’s journey began with a simple concept: connecting a trigger to an action. However, they quickly recognized a crucial insight from their users: the need for more complex workflows. This led to a significant product evolution around 2016, transforming Zapier from a mere integration platform into a robust workflow company.
But the evolution didn’t stop there. As users started building entire software solutions within Zapier, often bending existing tools like Google Sheets into makeshift databases, Zapier responded by adding essential components:
- Zapier Tables: Providing a backend storage layer.
- Zapier Interfaces: Offering front-end experience and forms.
This proactive adaptation to user needs has been a cornerstone of Zapier’s success.
The “Code Red” Moment: Embracing the AI Era 🚨
The emergence of models like GPT-3 marked a pivotal moment, prompting Zapier to declare a “code red.” Chris explains that this was a recognition of how AI would fundamentally change software development, aligning perfectly with Zapier’s mission of enabling people to build software without writing code. This realization spurred a reinvention of their business to support this new paradigm.
Agentic Workflows vs. Traditional Workflows: A New Level of Intelligence 🧠
The lines between traditional workflows and “agentic workflows” are blurring, but the core distinction lies in intelligence and adaptability.
- Traditional Workflows: These are linear and deterministic. A trigger initiates a sequence of actions, filters, and branching logic that executes the same way every time.
- Agentic Workflows: These agents are designed with three key capabilities:
- Access to Knowledge: They can retrieve and understand information.
- Ability to Take Action: They can perform tasks on your behalf.
- Ability to Reason: They can loop through a problem, adapt their course based on new information, and work towards a solution until it’s achieved.
Chris wisely notes that not every workflow needs to be agentic. There are still many valuable use cases where deterministic outputs are preferred.
Orchestration: The Art of Managing Your AI Workforce 👨💼
As we move towards more sophisticated AI applications, the term “orchestration” has become increasingly important. Chris likens it to hiring AI like an employee. It’s about how all these individual AI agents, working on your behalf, come together to achieve a larger objective.
Key aspects of orchestration include:
- Onboarding and Context: Providing agents with the necessary knowledge and context.
- Job Descriptions: Defining clear tasks and objectives for each agent.
- Interconnectivity: How agents are connected to your data and the tools they use daily.
- Proactive Work: Enabling agents to work on your behalf even when you’re not actively present.
At Zapier, they are currently orchestrating around 800 AI agents internally! These agents are triggered by schedules or real-world events, perform tasks, and report back for feedback.
APIs vs. MCPs: Bridging the Gap for Agents 🌐
Understanding how agents interact with tools is crucial. Chris clarifies the difference between APIs and MCPs (Modular Control Plane, though not explicitly defined in the transcript, the context implies a standardized way to expose tools):
- APIs: Allow an agent to be given documentation and then write software to query an app for data or take an action.
- MCPs: Provide a standardized way to define access to all available tools. These tools can be API calls or pieces of code. The key is a description that guides how and when the tool should be used, allowing agents to leverage pre-defined inputs without writing complex API calls from scratch each time.
Facing the Competition: OpenAI’s Agent Kit and Zapier’s Moat 🛡️
The rapid advancements in AI, particularly with releases like OpenAI’s Agent Kit, have led some to speculate about the future of platforms like Zapier. Chris addresses this head-on:
- OpenAI’s Agent Kit: Chris sees it as adding determinism and guardrails to agent building, a capability many agent builders already offer. He notes that it lacks Zapier’s robust trigger infrastructure and deep integration with API calls.
- Zapier’s Moat: Zapier’s competitive advantage has historically been its vast number of integrations and its ease of use. This creates a powerful feedback loop: more users lead to more integrations, attracting more partners.
- Future Differentiation: Moving forward, Zapier is focusing on the most use cases for automation and AI automation on their platform. They possess unique data insights into what’s truly shaping business transformation, a valuable asset for guiding users and enterprises.
Moving Upmarket: Serving the Enterprise 🏢
Zapier’s evolution from a product-led, self-serve model to serving enterprise clients involves a significant mindset shift.
- Self-Serve: Excels at understanding individual user needs and delivering solutions quickly.
- Enterprise: Requires a deeper understanding of organizational-level challenges and how automation can drive transformational value. This involves considering cultural and leadership changes alongside technological implementation.
To support this, Zapier has attached a sales function to the company, focusing on engineering value for large organizations.
Engineering Value: Empowering the “Closest to the Problem” 🛠️
A core belief at Zapier is that the people closest to the problem should be engineering the solution, even if they aren’t traditional engineers. Zapier’s tools empower individuals like HR leaders to solve their own workflow challenges, reducing the need for traditional handoffs and increasing self-sufficiency. This emphasis on ease of use is paramount for successful adoption and problem-solving.
Transformation vs. Adoption: Measuring Real Impact 📈
Chris beautifully distinguishes between:
- Adoption: Using new tools to do what you were already doing, but more efficiently. It’s about automating existing processes.
- Transformation: Doing something you couldn’t do before, fundamentally changing how work gets done. Automation and AI are key enablers of this.
This distinction is vital for executives who often use “transformation” as a buzzword. By breaking down value creation into these stages, businesses can better measure the impact of their automation and AI initiatives.
AI as a Teammate: Onboarding and Context Engineering 🤝
When thinking about AI as a teammate, the concept of context engineering is critical. Just as you onboard a human colleague, you need to provide AI agents with the necessary context. Chris’s personal use of a GPT agent to gather intel on podcast guests is a prime example of this.
The real transformative potential lies in extending this to team collaboration, where product, engineering, and design teams can build context together in shared spaces, rather than in silos.
Real-World Agent Examples at Zapier 💡
Chris shares compelling examples of how Zapier uses its own agents:
- Personal Productivity Agent: Runs daily at 8 AM, checks his calendar, identifies relevant documents using Glean, and provides a brief summary of external meeting guests.
- Product Launch Coordination Agent: During the lead-up to a co-pilot product relaunch, an agent monitored the weekly demos channel, analyzed transcripts and notes, and summarized key developments for the team.
He emphasizes the paradigm of “hiring an agent” over “creating an agent,” highlighting the onboarding and tool provision aspects.
AI Governance: Ensuring Safety and Trust in the Enterprise 🔒
For enterprises, AI governance is paramount. The core questions revolve around:
- Who is sending what data?
- What is the data?
- Where is it being sent?
Zapier addresses this through:
- Observability: Clearly answering the “who, what, where” questions.
- Access Control: Putting controls in place to enforce policies.
This leads to robust AI governance, ensuring the right policies are in place, enforced, and auditable.
Leadership and Commitment: The Key to AI Transformation 🚀
The journey to AI transformation requires strong leadership and genuine commitment. Zapier has:
- Declared a “code red” to acknowledge the significance of this moment.
- Appointed a Chief AI Officer (who is also the Chief People Officer) to prioritize AI initiatives.
- Engaged the executive team through “AI show-and-tell” sessions, encouraging hands-on experience and a deep understanding of AI’s capabilities and limitations.
Chris stresses that leaders must be hands-on with the tools to foster authenticity and drive meaningful change. It’s not just about sending memos; it’s about building and experimenting with AI.
The Future is Automated, and It’s Here! ✨
Zapier’s journey exemplifies how a company can not only adapt to technological shifts but also lead them. By focusing on user needs, embracing innovation, and strategically orchestrating AI agents, they are building the future of work, empowering businesses of all sizes to achieve unprecedented levels of efficiency and transformation.
What are your thoughts on agentic workflows and AI orchestration? Share your insights in the comments below! 👇