AI is going to kill {Company X} - Will MCP zap Zapier

🧐 Chegg’s stock plunged 89% due to AI, while MCP threatens Zapier’s 8,000+ integrations, pushing it to adapt or decline.

A favorite topic for online hot takes these days is how some businesses are about to be killed by AI. Legacy, in this context, is a business that existed before ChatGPT. [1]

Some companies are clearly already in trouble, Chegg, purveyor of study guides, has been justifiably crushed in the market because ChatGPT et al clearly replaced a huge part of their market. As a result, Chegg’s stock price is down 89% in the last year, ouch.

• Chegg’s stock performance of the past year, the big drop in early 2024 was after they announced on their earnings call how LLMs were already impacting revenue

An example that piqued my interest last week was the exclamation that companies like Zapier are doomed because of the Model Context Protocol (MCP). I’m not convinced.

First, here is a quick definition from Anthropic’s announcement about MCP.

It provides a universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol. The result is a simpler, more reliable way to give AI systems access to the data they need.

I work in the data integration business and have for some time, so I have no doubt about the importance of developing standards for interoperability. MCP has good early traction. Assuming that continues, it will clearly have a big impact on how AI can be deployed.

Back to Zapier they are an integration platform that connects different SaaS systems via their APIs to sync data directly. For example, you might use it to connect a CRM like Hubspot with a call recording tool to sync a summary from every call to the relevant deal.

An example of a Zapier flow moving data from a Facebook form to trigger an email in Gmail and a Slack alert

Pre-AI, Zapier was a darling of the SaaS world, scaled to $310+ million in revenue with just $2.3 million in funding.

So, is MCP going to kill Zapier? No, or, at the very least, it’s not a foregone conclusion. Here’s why.

If MCP was quickly adopted by all of the apps that Zapier helps to integrate, 8,000+, they would definitely be in trouble. That is unlikely, partly because of inertia; it takes time for companies to update their API interfaces, and partly because it may not be in the apps’ interest to make their data available to AI in general vs. offering it as a part of their app (more on that later).

That leaves us with a situation where Zapier is actually extremely well-positioned for MCP and AI.

They already have all of the integrations and an architecture for connecting across them. In principle, they just need to implement MCP once at the central layer. If Zapier doesn’t do this, then it's a failure of execution. They may already be leveraging MCP; both Zapier and competitor n8n have pages highlighting their integrations with Claude (though neither specifically mentions MCP).

Will Zapier get replaced by AI? Maybe. It’s possible that at a lower level, beyond MCP, AI gets so good at writing code that you can just ask for integration with any arbitrary API interface and it will deliver one. Given that OpenAI uses a 3rd party SaaS tool for their ETL, we are definitely not at that point yet. In my mind, Zapier and others in their category are well positioned, at least for the short to mid-term.

If we take a step back from Zapier, it’s worth looking at the dynamics at play. There’s been plenty of noise about the end of SaaS. Perhaps most notably, Klarna claimed to be replacing Salesforce and Workday with in-house solutions created with the support of AI. Just 6 months later, Klarna’s CEO walked back those comments and noted that a more likely trend is that there is consolidation in the market, which he’s probably correct about.

A classic model of why new startups win against incumbents is they: a.) move faster b.) are built for the new paradigm (e.g. cloud vs on-prem) and/or c.) not limited by the innovator's dilemma. While these are all relevant to AI, the dynamics are different. There is also a new factor, data.

Let’s start with data. All the major models are trained on the internet and have more or less exhausted the supply of publicly available data. If you have proprietary data sources, you have a huge leg up. Incumbents have the data, startups do not.

Revisiting Salesforce, they have all of the data of their 150,000 customers to train their Agentforce services on. Attio, an AI-first CRM, has a much smaller customer base and is presumably just wrapping standard off-the-shelf models. Which is more defensible?

On the innovators dilemma, which Kodak is the canonical example. They invented the technology for digital cameras but there was no business model for digital photography that would have ever replaced the revenue from film (other than Kodak developing the iphone).

There are certainly sectors where incumbents will fall victim to the innovators dilemma, e.g. law firms whose revenue comes from billable hours, but for a lot of software players, AI is additive. They can augment their services and charge more or deliver better outcomes, assisted by their data advantages.

Google Search is perhaps the biggest revenue stream exposed to an innovators dilemma with AI. Some have made the case for why they will lose but early indicators are that Google Search is doing just fine, in part because AI chat is solving new problems rather than replacing Google, but also because it’s still tiny in comparison.

So yes, AI startups will unquestionably move faster and build specifically around AI as a paradigm but in a lot of sectors, incumbents will be able to leverage their data advantage and existing customer bases to win out over startups.

Regardless of whether startups or incumbents win. I know data is critical to success in an AI-first world. It’s going to be an interesting few years upcoming.

Warmly,
Paul Dudley

Footnotes:
[1] Have we entered a new era? Is it no longer 2025 AD but rather 3 AG (After GPT).