Pod Insights The Agile Daily Standup - AgileDad The AI TPM Landscape About this episode The AI TPM Landscape
If you search for "AI Technical Program Manager career advice," you'll find two things: job postings and generic reassurance that "your skills transfer." What you won't find is anyone mapping the actual landscape showing which companies expect what, and where your existing experience actually gets you.
I've spent the last three years at a cybersecurity company during its AI transformation, watching how AI products actually ship. I've analyzed job descriptions across dozens of companies spanning three tiers (Frontier AI, AI-Applied-to-Business, and AI-Powered Applications).
AI TPM roles vary more than the job titles suggest. Some require deep fluency in model development. Others require working knowledge of how to ship AI-powered products without building the models yourself.
This distinction matters because it defines the depth of your technical pivot. It determines whether you need to master the 'physics' of model development, the 'orchestration' of AI-powered systems, or the strategic integration of AI into existing workflows.
Transcript 0:15 Speaker 00 So I have another one that I've recently been asked to review and this one came from LinkedIn as well. It was one of my LinkedIn subscribers who recommended that I review an article. So I'm going to go ahead and deep dive into this. I'm gonna give you feedback as I go, but this is probably going to take the entire 15 minutes. So I'm not going to give you any upfront preview. So here we go. If you search for AI technical program manager, career advice, you'll find that two, you'll find two things. Job postings and generic reassurance that your skills transfer. What you won't find is anyone mapping the actual landscape, showing which companies expect what and where your existing experience actually gets you.
0:56 Speaker 00 I have spent the last three years at a cybersecurity company during its AI transformation, watching how AI products actually ship. I've analyzed job descriptions across dozens of companies, including for, including three tiers. Frontier AI, AI applied to business and AI powered applications. AI TPM roles vary more than the job title suggests. Some require deep fluency in model development, others require working knowledge enough and working knowledge of how to ship AI powered products without building models yourself. This distinction matters because it defines the depth of your technical pivot. It determines whether you need to master the physics of model development and the orchestration of AI powered systems
1:40 Speaker 00 or the strategic integration of AI into existing workflows. So let's pause there for a second. So I think this is going to be a deep dive into the AI technical project manager role. And what I can tell you is this exists, so let's continue. Where are the AI jobs actually are? According to Menlo Ventures 2025 State of General AI Report, enterprise AI spend on software and services hit 37 billion in 2025, up 3.2x year over year. This is the fastest scaling software category in history. That money flows to three types of work, building foundation models, applying AI to business problems, and integrating AI into existing products. Let's pause there.
2:25 Speaker 00 So I can tell you that this is something I definitely see, and it is true. The part where we integrate is applying AI to help solve business problems. We do teach a lot about the others, but that's really our forte at Agile Dad.
2:38 Speaker 00 But spending categories don't map cleanly to TPM roles. A tier one company might need both deep model expertise and productization. TPM, oh, they might need both of those types of TPMs, that makes sense. A tier two company might need everything the ML pipeline suggests, a specialist to integration experts. The point is where the opportunity sits, not where the dollars flow. Seventy-six percent of all solutions are now purchased rather than built, up from fifty-three percent in 2024. Companies are buying AI capabilities, not building them from scratch. And that means the majority of AI TPM roles are about integrating and operationalizing AI, not training foundation models how to do it.
3:22 Speaker 00 But what does it actually take to succeed as an AI TPM? It starts with understanding what AI fluency really means. So I define AI fluency for TPMs as the ability to understand AI ML concepts well enough to facilitate technical decisions, ask the right questions, and translate between ML engineers and business stakeholders. You're not building models. You're making sure the right models ship. The required level of AI fluency varies by tier. So tier one is what we call frontier AI. This is a researcher, foundation level model, so Anthropic, OpenAI, DeepMind, etc.
4:05 Speaker 00 So the AI fluency required there is very deep. You must understand model training, evals, compute, you have to understand all those things. Then there's tier two, which is AI applied to business. This is kind of where I live. Product AI teams, Meta Ads Raking, Stripe, Risk, Netflix Recommendations. This is understanding ML pipelines, A/B testing, model metrics, prompt engineering, responsible AI, and agent evaluation. This is a great place to sit. And then tier three is AI-powered applications. AI feature integration, ServiceNow, Workday, Atlassian, etc. It's conversational.
4:44 Speaker 00 Evaluate AI solutions, define quality, define quality gates, discuss features, substantivity, etc.
4:54 Speaker 00 So, one clarification, a company's tier and a role's tier are not the same. Meta built Llama, but the Ads ranking TPM role is tier two. Google has DeepMind, but the most cases across the company span a full range from tier one research programs to tier two product AI to tier three platform work. The question isn't, is the, is the, is this an AI company? It is, what does the specific role within the company require on the AI scale? That makes a lot more sense. So if you're looking for tier one, frontier API with deep fluency. These are the roles of companies that push the boundaries on AI's capabilities.
5:34 Speaker 00 So they're computing, capacity planning, model evaluation, research to production handoffs, infrastructure management, AI safety policies, all those good things. From actual Anthropic J Ds. Drive execution of complete capacity planning and allocation, orchestrating complex model launches, contributing to initiatives related to security commitments and Anthropic's responsible scaling policy. This is fundamentally different work. If you don't know what RLHF means, or why compute allocation matters, you need to close that gap before these roles are even realistic. Okay. So we're about a third of the way through the article, and what I can tell you is this is really interesting.
6:18 Speaker 00 I like this three-tiered approach. Let's continue. Tier two is AI applied to business or work influency related. These roles apply to AI, ML to core business problems. TPMs manage ML-powered products, but don't work on foundational model development. So examples they give are ML pipeline coordination, feature engineering, model training, deployment, A B testing at scale. Experimentation, frameworks, statistical significance, model performance metrics, precision, recall, latency, and business KPIs. Cross-functional execution. This is the core TPM job. Roadmaps, dependencies, stakeholder alignment, etc.
7:03 Speaker 00 So from a meta ads ranking TPM JD, he says, partner with monetization engineering and ML ranking teams, develop strategies on how to measure and improve revenue, scale ads and recommendation technologies. So I like that. Most of the standard TPM work roadmap stakeholder management execution, the AI specific piece is understanding ML pipelines well enough to make sound program decisions. In practice, the fluency that matters on tier two comes down to defining evaluation criteria that both ML engineers and sales leadership can align on, facilitating trade off decisions between accuracy and latency
7:48 Speaker 00 and building quality gates that predict production success. The ML team owns the models, the TPM owns the process that gets them shipped. And then there's tier three AI powered applications, conversation fluency required. These roles are focused on adding AI features to existing products. So TPMs manage AI enhanced programs, but the core work is traditional still. So there's feature integration, vendor management, product launches, platform launches, et cetera. From a service now senior staff TPM JD, drive technical programs across platform engineering, partner with engineering leadership on roadmap execution, establish operating cadences and governance.
8:32 Speaker 00 And from and this is traditional TPM work with an AI layer. You need to understand what AI features do, evaluate vendor solutions and define quality gates, but you're not managing the entire project or pipeline. Things I wish I would have known for tier three, you probably already qualify. You need enough AI fluency to evaluate solutions and define quality gates, not ML expertise. A few weeks of focused learning and one hands on project and you're there. All right, for tier two, AI applied to business. That one's a little more.
9:07 Speaker 00 Yeah, most of what's needed, ML Pipeline, Understanding, Evaluation Metrics, and AV Testing at Scale are learnable in three to six months with a little bit of focused effort. And then for Tier One, which is Frontier AI, this requires a genuine investment of, plan for at least six to 12 months of serious coursework and hands-on projects with adjacent experience. This barrier depends entirely on which tier you're targeting. And here's the good news, most AI TPM jobs are for Tier Two and Tier Three Frontier companies, or for Tier Two and Tier Three. Frontier companies are growing fast but remain relatively small, like Anthropic and OpenAI, each have over a few thousand employees at this point.
9:47 Speaker 00 The real volume is the companies applying AI to business problems, or adding AI features to existing products. That's where the traditional TPM skills translate most directly. It doesn't mean Tier One is out of reach, it just means you have an options while you're moving in towards it. So where's all this AI stuff heading? That's the big $6 million question. My honest answer, no one knows exactly. No one's gonna know how it shakes out, but I'm watching three forces. One is fine tuning divide. Generic AI APIs lead to generic products. Companies serious about moat building will fine tune the proprietary data. This shifts them from Tier Three to Tier Two.
10:32 Speaker 00 And if you're one of these companies, your fluency requirements have just changed. Okay, let's go to the next one, which is the orchestration problem. As AI agents proliferate, the hardest work won't be turning individual models. It'll be architecting how they work together. Think less ML Pipeline TPM and more AI Systems Architect. This could become its own tier. And then there's the ROI correction. After years of AI experiments, boards are demanding results. Some complex tier two projects will get scrapped in favor of simpler integrations that actually move the needle. The question becomes, does this make us money?
11:14 Speaker 00 Instead of the current, how sophisticated is the AI technology we're trying to use? My bet, TPMS who thrive will be architects of AI powered systems, not just integrated AI services. But regardless of which prediction plays out, the TPM fundamentals, shipping, cross functional alignment, knowing when to cut scope, become more valuable, not less valuable. The models will evolve, but the need for someone to cross align humans and align humans in general, it just... The need for someone to align humans around them won't. It just won't work.
11:57 Speaker 00 So I hope you understand this high level and I hope you see where you are in the tier one, tier two and tier three. This article was well written. This article was also very, how should I put it? Very flat when it came to emotion, which I know is fine. But the truth is, this TPM landscape is definitely real. There are technical project managers out there. I call them technical analysts, you've called TPMS. But these people do exist. And I would argue that right now their role is evolutionary. And I hope that over time it will become more standardized and more formal. Because I can tell you right now that this is definitely going to be something that will work out in your behalf. So that's going to do it for today.
12:42 Speaker 00 I enjoyed the article. Thanks for sending it over. If you have a topic you want us to discuss, learn more at agiledad.com. I'd love to hear from you. And as always, I encourage you to stay healthy, stay well and stay agile, my friends. Until next time, do take care. Bye now.
Previous Scrum Is Dead - Here Is What Killed It! Next Which Leadership Pattern Shows Up Under Pressure? - Mike Cohn Page updated November 6, 2026 · transcript and summary are machine-generated