By Tela Mathias, Chief Mad Scientist at PhoenixTeam

We’ve been in the lab for the past few weeks tackling the agent problem and today it really started to connect. I’ll start with a reminder about what an agent is. The foundation of agentic AI is its ability to reason. An AI agent is characterized by the ability to perceive and understand context, reason about a problem, plan and take action, and use tools.

The agent problem, of course, is understanding and rapidly deploying… agents. Why does this matter? This matters because it is a huge unlock in time to value. When we master agents, we eliminate (or at the very least drastically reduce) reliance on coding for solution delivery. And that means we can move at a blistering pace.

We believe that in the AI future, both the front end and the backend of solution delivery are commoditized. There is no moat in foundation models (the “backend”), and anything we can image, we can build (the “frontend”). So what, then, are the differentiators? Original thinking and vertical industry expertise. We believe that in the future (which is happening now, by the way), there is a new formula for differentiation.

Differentiation = Original Thinking + Domain Expertise + Fast Scale

Considering this, we’ve been in the lab working on going “concept to cash” (idea to production value realization), in five days. Imagine if you could have an idea on Monday, and see it through to production by Friday? Wouldn’t that be amazing? I know this is possible. I can’t say we’ve got it quite down to five days, but we are not that far off. What if we could pick an agentic use case from a menu, and have that use case up and running within a week? For that matter, what if it was two weeks? Certainly, way better than how long it takes to get stuff into production in a more classic approach.

I still love classic, don’t get me wrong. We have plenty of customers where we work with AI accelerators to deliver software using traditional agile scrum, with continuous discovery and continuous delivery. That is very much still a thing. But there is another way too, and a way that will become a massive differentiator for those clients boldly willing to go where no one has gone before.

So you are thinking – “yes, let’s go!”. Well, hold on a minute. The tooling isn’t there yet. Where it is there, it doesn’t scale. The accelerators are nascent, and connecting the front end to the back end is still a long way off from “push button get software”. But it’s coming, and it’s coming fast. Things like model context protocol (MCP) will ultimately make building agents like building with Legos. The pieces are all there and you just snap them together. Voice agents will create the ability to do customer discovery at scale in, like, a morning. Coding agents are already accelerating development.

The human side of this is interesting and has to be cared for. There is a lot of fear. Jobs are changing. The things we have to know how to do come with a really steep learning curve and not all of us are ready. I talk often now about what this means to my kids. This generation will be the last that was born before ChatGPT. They are also the unique set of children that had a critical time in their lives and learning utterly disrupted by covid.  They will engage with the world and learn in fundamentally different ways. The education systems are not changing fast enough, which puts the burden on parents. We are the ones that have to help our children adapt. And we don’t even know what that means so it’s a bit of the blind leading the blind.

The agents are here, and they are coming for our kids.

I have to just acknowledge how scary that is. I do worry. I worry about my oldest at 22, and my youngest at seven (and everyone in between, my nine and eleven year olds). But we really don’t have a choice. The future is coming and the only way out is through.

For existing professionals, it is also scary out there. Will I be replaced? They way I mastered my craft is not the way anymore. Is my craft even still relevant? The reason I know anything about subservicing is because I spent three years photocopying checks in the second subbasement of HUD in conjunction with litigation support for a federal case against a Ginnie Mae master subservicer whose assets were seized by FDIC. That was 1998 and even then it was old (they were raided by the FBI in the early 1980s).

As much as I came to hate this job, gosh did I learn a lot. Precision. Attention to detail. Escrow account commingling. 11710A monthly reporting. A goldmine really, still useful today. This is simply not a job anyone will do in the AI future. Agents will sift through massive quantities of structured and unstructured and structured data, decide what needs to be done, and do it. How will new professionals learn these things? Does it even matter? What do they have to learn?

I had a fascinating conversation with a senior executive talking about interns last week. They have their interns learning to do things “the classic way”. Stare and compare. Good old fashion math and analysis. This is a valid approach, but it is not the one we are taking. We have our interns living in the AI future. We are paying them to learn, explore, try, and fail fast – all using “the now way”. I have no idea if this is “right” but it’s the path we chose. For now?

I choose to see the sunny side of this, and really I try to see the sunny side of all things. The alternative is depression and despair, and I don’t want to live like that. We all survived the internet, and we will survive this. Those of us who are out in front will pave the way for everyone else. It’s bumpy and painful but it’s better than being bored.

‍