On this episode of Innovators, I spoke with Jason Ambrose of People.ai about what “agentic AI” actually means, why sales data is messier than most people think, and why blindly trusting large language models is a mistake. People.ai has been around long enough to see multiple waves of enterprise software come and go. Now it’s repositioning itself squarely in the agent era. Most CRM systems tell you what was entered. They don’t tell you what’s actually happening. People.ai takes a different approach. Instead of relying on manual updates, their AI analyzes the communications that define modern sales, emails, Slack messages, meetings, chat transcripts. The system maps that activity to accounts, contacts, and opportunities. That sounds straightforward until you scale it up. If you’re a startup selling to a small business, maybe one salesperson is talking to one buyer about one product. That’s simple. But when Microsoft sells to Verizon, you might have dozens of people on both sides, across legal, technical, procurement, and executive roles. Conversations happen everywhere. Mapping that complexity into a clean CRM record is hard. That’s where People.ai claims it shines. It uses its own AI models, trained on billions of transactions, to reconstruct what’s really going on inside a sales organization. What Is an Agent, Really? We talked about the shift from chatbots to agents. A chatbot answers a question. An agent has an objective. Jason framed it in terms of business process automation. Old-school automation works when the logic is predictable. If this, then that. Stay inside one system, follow a defined workflow. Agents step in when reasoning is required. They cut across systems. They pursue a goal. They have to decide what to do next. But that only works if they’re plugged into real expertise. Jason made a useful distinction. Public LLMs are trained on public data. Enterprise expertise lives in private systems. If you want an agent to act intelligently inside a company, it needs access to proprietary data. That’s a big trust ask. You’re effectively saying, “Let our AI read your emails.” That’s not a small decision. Avoiding “Build Trust With Stakeholders” Anyone who has used a generic LLM for business advice has seen the problem. You ask for guidance and you get vague platitudes. “Build trust.” “Accelerate the deal.” “Engage the customer.” That’s not actionable. Jason argues that this is where expert agents come in. Instead of spitting out generalized advice, they ground recommendations in specific deal data. Who hasn’t responded in three weeks? Which technical blocker hasn’t been addressed? Where did the last conversation stall? Without that grounding, AI defaults to corporate fortune-cookie language. The Capital Markets Reality We also touched on fundraising. SaaS is being repriced. Public markets adjusted first, and private markets followed. Companies that once enjoyed premium multiples are now being reevaluated in light of AI disruption. Capital is flowing into AI-native plays. If you look like “just another SaaS company,” you need a credible AI story. If you genuinely sit at the center of AI transformation, you’re in a stronger position. People.ai is not currently raising, but Jason sees the shift clearly. The market is asking who is being disrupted by AI and who is using it to build something new. Is AI Replacing Jobs? It’s the obvious question. Jason’s take was pragmatic. Technology changes work. It always has. He remembers the early days of the web and the anxiety that came with it. Some jobs disappear. Most jobs change. His line stuck with me: people should work with people, and let AI do the rest. Sales, at its core, is still about relationships. AI can summarize, surface risks, and suggest next steps. It can’t replace trust, empathy, or judgment. At least not yet. If you’re in sales and you haven’t started using AI, Jason’s advice is simple. Start. Use ChatGPT, Claude, Gemini, whatever tool you prefer. Have it rewrite emails. Summarize meeting notes. Draft follow-ups. But don’t copy and paste. He pointed out something many executives are quietly thinking: if you send a clearly AI-generated email without tailoring it, you’re signaling that you didn’t invest the time. And if you didn’t invest the time, why should the recipient? AI can amplify your work. It can’t replace the part that makes you human. People.ai is betting that the future of sales is agentic, cross-system, and grounded in real communications data. Not just dashboards, but reasoning systems that understand what’s actually happening inside complex deals. Whether you buy that vision or not, one thing is clear. The next phase of enterprise AI won’t be about novelty. It will be about integration, trust, and measurable outcomes. And that’s a much harder problem than writing clever emails. TRANSCRIPT Welcome back to the Innovators show about amazing people doing very cool things. I’m John Biggs. Today on the show we have Jason Ambrose from People.ai. It’s agentic and it’s for sales teams, but why don’t you add to that, Jason, welcome. Jason Ambrose (00:24.482) Yeah, thanks, John. So what People.ai does is our AI figures out what’s happening in a sales organization by looking at the communications between your field and your customers. So we analyze emails, chat transcripts, meetings, Slack messages, and the like to turn that beyond just the data to what’s actually happening. How does that how do you find the answers of what’s happening in the organization? And we provide that either to humans or to agents. So that’s been our big shift this year is to realize that the stuff that we were doing for humans in CRM is also very relevant when you have agents trying to figure out what’s happening in sales. John Biggs (01:04.094) So let’s explain agents to folks who might not even understand what’s going on. So the idea originally was that you had a chat bot. You asked it something, and it responded to you. But now we’re talking about agents, which are supposed to be autonomous to a degree. So how do you guys describe those, and how do you use them? Jason Ambrose (01:24.086) Yeah. And hey, look, you know, I may not have everything right on this too, but at least the way that I think about it is maybe starting from a business process automation, right? So, you know, for, for periods of time when we had predictable workflows and we knew, you know, sort of if then else, there’s not thinking that happens there, but we could automate work if that had to happen. John Biggs (01:28.188) Mm-hmm. Yeah. Jason Ambrose (01:48.302) In the case of agents, that now becomes something where they have some chain of thought, they have some reasoning. So they know they have an objective or a purpose that they’re trying to work through. They have to figure out how to get that done. So when there’s a little bit more, you know, thinking, reasoning that needs to happen to figure out how to get that objective, that suits an agent. What I think we’re seeing with customers is they’re figuring out how to unlock that for work that needs to get done across a lot of different systems, right? So, know, BPA, business process automation, or what you have in your workflow tools that tends to say within the silo of a system from data to business roles to presentation layer to humans. When you start to cut across the systems, that’s where there’s been big opportunities for agents. John Biggs (02:37.214) So in this particular case, you guys are focusing on sales leads, that sort of thing. So you basically take every single data point that you have and say, this person, I don’t know, emailed you two weeks ago and also was tweeting this and is interested in this. So why don’t you give him a ring? Is that generally how it works, or what’s the? Jason Ambrose (02:56.376) That’s yeah, that’s really close. Yeah. I think the difference is, you know, let’s think about two different types of selling, right? you could be a startup and you’re selling to a small business. That’s, know, pretty much one buyer. So, you know, if you think about it in the context of CRM, you’ve got one salesperson. You’re selling to one buyer at one account and you’re selling one product that that is pretty simple to figure out, right? Where it gets more complicated is if you’re. Microsoft selling to Verizon just to pick two big companies. You might have 30 or 40 people or more on the Microsoft side. You might have 30 or 40 people on the Verizon side answering different technical questions, having different conversations about different elements of your business relationship and how you match those activities to records in CRM that represent, you know, here’s a person that we’re talking to, here’s the account that we’re talking to. you know, here’s the specific sales opportunity that becomes really hard to do properly. And that’s, that’s where we, that’s where we shine. And that’s where we have, you know, pretty large customers like Red Hat, Verizon as a customer and some others. John Biggs (04:09.712) Would you be able to still do this without AI? Would this exist if we didn’t have this kind of, I don’t know, synthesis, right? Jason Ambrose (04:15.798) It would be really difficult, right? So we have our own AI that’s applied to do the math to figure this out. And it’s learned from looking at billions of transactions over the years, right? The second piece, I think, is how you integrate or interface with other systems. So you mentioned the chat interface. So a human does want to do that, right? So we put this alongside sales opportunity record. If you want to get the full story, you can ask the chat bot, are the risks in these deals or what’s happening in this account? Now with MCP, that same type of interaction can happen from an agent to our system.
