Why MCP Protocol Alone Won't Fix AI Agents | Ayan Barua
MTS
October 8, 2026
Ampersand CEO Ayan Barua breaks down why traditional unified APIs and Model Context Protocol (MCP) aren't enough to connect AI agents to legacy enterprise systems, and how real-time translation layers are solving the reliability bottleneck. Turn ideas into software people love.
Ayan Barua
MCP is another protocol, and sometimes the protocol is not enough. There's a lot more orchestration work that you have to do.
So MCP makes tool calling easy, but you are literally thinking of tool calling as the only thing that you will call. So MCP is a part of the stack, not the solution. You need a lot more tooling on top of it to make it work in the enterprise.
Sophia
Hello, everyone, and welcome back to MTS. Today, we are joined by Ayan Barua, who is the co-founder and CEO of Ampersand. Ampersand is a developer platform that helps AI agents connect to enterprise software like Salesforce and HubSpot. Yesterday, they just announced their 15 million Series A, led by Bessemer Venture Partners. So Ayan, welcome to MTS.
Ayan Barua
Hey, thanks for having me.
Sophia
Yeah, thank you for being here. Wanted to talk about this big announcement that you guys had. So you guys just announced this $15 million Series A. What's the core problem that you guys are trying to solve?
Ayan Barua
Yeah, the core problem is essentially enterprise AI has an integration problem.
So there's a lot of new products that are coming out, whether it's from the labs or from eGenTech software products. Now, all of them fit into the AI transformation budget. And there are $10 trillion of work that is locked into legacy systems. And the legacy systems will be systems of record, like Salesforce or Opspart or NetSpeed or Oracle. So new software will need to speak really nicely with previous generation software. So we are building a general purpose translation layer for the developers of new software when they offer their agents to their customers, their customers connecting to these legacy systems that's syncing happens very naturally. So that's kind of, we're building the rails for that.
I'm excited to take it to the next level now.
Sophia
Yeah, no, super interesting. And I know that you guys have specifically said that traditional unified API model doesn't totally work because they can understand more complex data structures. And so why don't I just get your takes from this like tech stack layer? Like what is now obsolete? How is that whole thing changing?
Ayan Barua
This integration problem that I'm talking about is just not an AI problem. It has been existing ever since business software existed, right? So the newer generation products work more real time. The newer generation products will also be accessing all parts of the stack, right? So we've had multiple iterations of integration products. In a unified API, it was the last iteration.
And it took a world view of a standardized model. And then you could code against it. With AI, I think the code has become much cheaper to do. It's basically understanding the customer implementation per customer, right? So we are taking a different approach. We are basically saying that you don't need a unified API model, or you don't need a common model. You should have one model that works for yourself as a software builder. And then your customer can connect their stuff to your model, right? So it can happen per customer.
So if you see the world, there's a lot of FDEs right now in the world, right? So they are custom building stuff in the enterprise. So if you do too much of that, it's a service business, right? You need to productize it. So we are building these set of platforms where if you are offering these to the enterprise as a product, you don't need to custom implement them per customer.
Sophia
Yeah, definitely.
I know one of your big focuses right now is on real time enterprise data. And so tell me a little bit what that means. Why are enterprises asking for that? And just share what you've learned working with teams like Salesforce and HubSpot, and what are the core things that they're looking for?
Ayan Barua
I think customers just, with EI, the promise is now, right? Like I want value to be delivered to me now, not like three months later or six months later, right? So I think the real-time-ness is coming from the fact that generally the world has become more impatient. They want things to be today, not tomorrow, right? So we are looking at a faster pace of software deployment, faster pace of adoption. ERP transformation cannot take two years, you know, like people want that to be in two months, right? So if you cut down the overall deployment of software by 10 times or 100 times, then every part of the stack needs to move, you know, much faster. So the real-time-ness is really like a function of that, that if there is five seconds that you can shave off in the request response, you know, then we should just go do it, right? So this is what we're seeing, like our customers, we sell to software developers, their customers want, you know, faster deployment, faster reaction times, right? So we are infrastructure for doing that.
Read the other 7 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000793813756