Derek MillsJPMorgan's AI Mandate Isn't Optional and Your Software Team Knows It
March 27, 2026
I want to talk about what JPMorgan is actually doing, because most coverage has missed the part that matters to anyone running a business that depends on a software team. And right now, almost every B2B company depends on a software team.
Here's the short version: JPMorgan Chase is spending $19.8 billion on technology in 2026. Not as a gesture. Not as a PR move about innovation. They have built an internal AI platform called LLM Suite that runs models from OpenAI and Anthropic, updates every eight weeks, and has been deployed to over 200,000 employees. More than 40,000 of their software engineers are using AI coding assistants right now. Jamie Dimon has publicly acknowledged the bank has already displaced workers from AI and is redeploying them into other roles. The chief analytics officer's stated end goal is a future where "every employee will have their own personalized AI assistant, every process is powered by AI agents, and every client experience has an AI concierge."
That is not a vision statement. That is an operational roadmap with a budget larger than most countries' defense spending.
I've been watching this story build for a while. Tory kept telling me it was "just another big bank doing tech theater" - this was a few weeks ago, right around the time his car got repossessed, so his judgment on large capital commitments was maybe not at its sharpest. But he's not entirely wrong that most enterprise AI announcements deserve skepticism. JPMorgan is different. The numbers are too specific and the timeline too concrete to wave away.
What JPMorgan Actually Built (And Why It Matters That It's Theirs)
The LLM Suite story is the one people keep glossing over. This isn't a company that signed a Microsoft Copilot deal and called it a transformation. JPMorgan built their own internal platform, made it model-agnostic so they could swap between AI providers, and pushed it to employees through what their chief analytics officer Derek Waldron described as an opt-in approach that created "healthy competition, driving viral adoption." LLM Suite went from launch to 200,000 users in eight months. That is a faster internal adoption rate than most SaaS companies achieve with paid enterprise contracts.
The reason they built internally rather than buying off the shelf is the same reason it's terrifying for the rest of the software industry: data control. When you're handling $4 trillion in assets under management and 84 million US customers, you cannot hand your data to a third-party vendor and hope for the best. So they built the moat themselves. The AI is being fed proprietary data from every major business unit, updated on a fixed cadence, and locked behind their own governance framework. The more they feed it, the more valuable it becomes, and the harder it is for a competitor to replicate.
Investment bankers are generating five-page pitch decks in 30 seconds. Wealth advisors are finding research information up to 95% faster. The wealth management division reported gross sales rose 20% year-over-year between 2023 and 2024, with AI-driven tooling cited as a contributing factor. These aren't hypothetical gains. They're in earnings calls and investor day presentations with specific numbers attached.
This reminds me of something Linda said last week - she and Gerald were talking about how their bank suddenly got much better at recommending things. She didn't know why it felt different. Now I know why.
The Part About Software Engineers Specifically
Here's where I want to be direct, because the software industry has spent two years telling itself that AI coding tools are for amateurs and that senior engineers are safe. JPMorgan just stress-tested that argument at scale and the results aren't comforting.
Over 40,000 software engineers at one institution are using AI coding assistants right now. Not because they were forced to in some top-down "use this or else" way, but because the bank made the tools available, measured the productivity gains, and made the case from the results. CFO Jeremy Barnum talked about it in a way that felt almost startled - he said he'd been "vibe coding" personally and found it "actually pretty amazing," then confirmed that trained professional engineers were also getting significant efficiency gains. When a CFO is casually talking about vibe coding in an investor day presentation, the cultural shift has already happened.
The bank is tracking use cases and Barnum confirmed they've doubled generative AI use cases in a single year, with the biggest focus landing on customer service and software engineering specifically. This is not the support functions or the back office dipping its toes in. This is the core engineering org.
And then there's the workforce math that nobody wants to say out loud: operations staff are projected to fall at least 10%. Dimon himself said JPMorgan will likely have fewer total employees in five years. He's not hiding it - he's explicitly calling on governments to build retraining infrastructure before the displacement happens rather than in response to it. That kind of candor from the CEO of the world's largest bank by market cap doesn't happen unless the internal data is very clear about what's coming.
I keep thinking about the scene in The Last Jedi where the Resistance is just bleeding ships on that slow-motion chase through space, and everyone watching thinks nothing is happening. Then suddenly almost everyone is gone. That's what this labor transition feels like for software-adjacent roles. Slow, then all at once. Most people keep saying The Last Jedi has no stakes. I've made this point eleven times this week. Tory asked me why I care so much. I care because the slow tension IS the stakes - same as what's happening in tech right now.
Why This Is Your Problem Even If You're Not a Bank
The companies that move slowest on AI integration aren't the ones that will get disrupted by AI directly. They're the ones that will get outcompeted by companies that did move - and lost on speed, cost, and output quality simultaneously.
McKinsey estimates there's roughly $700 billion in potential cost savings industry-wide from AI adoption in banking alone. Their analysts think much of that will get competed away to customers, and that AI pioneers could see four-point increases in return on tangible equity versus slow movers. Four points is a massive spread in an industry where margins are measured in fractions. Translate that logic to SaaS, professional services, logistics, manufacturing - any sector where knowledge work is the product - and the same dynamic applies.
JPMorgan's technology budget is now approximately 10% of their total revenue. Dimon called that figure unremarkable - he said most companies spend more than that proportionally. Think about what that means for B2B software buyers right now. Your vendors are feeling this pressure. The ones that figure it out will pull away from the ones that don't. If you're evaluating platforms and comparing features, you should also be asking what percentage of revenue they're putting toward AI development and whether they're building internal capability or just bolting on someone else's API. The answer tells you a lot about where they'll be in three years.
Stephanie, our resident person-who-has-never-once-worried-about-a-budget, glanced at JPMorgan's $19.8 billion tech number and said "oh that's not that much for a bank, is it?" and walked away. Which is a very Stephanie reaction. But she's accidentally correct that the number is less shocking in context than it sounds. Bank of America is spending $12 billion. Goldman Sachs committed $6 billion. The whole sector is moving. JPMorgan is just moving fastest and being most public about it.
For anyone managing revenue operations or sales infrastructure, the implication here is that the AI tools in your stack are going to start getting compared - by your buyers, your board, your competitors - against what a maximally AI-integrated organization looks like. That bar just got raised by a company with 317,000 employees and a two-thousand person dedicated AI team. The revenue intelligence landscape is already reshaping around that standard, and the sales automation category is moving in the same direction.
The Mandate Part Is What Everyone Is Underreacting To
When people say "JPMorgan's AI mandate," they mean two things. First, the internal directive: use AI tools, get trained through the "AI Made Easy" program, integrate this into your daily work. Second, and more importantly, the external signal: this is what competitive looks like now.
The bank set up a firmwide Chief Data and Analytics Office specifically to govern AI adoption - not to study it, but to run it. They have over 450 AI use cases in production across the business. They're now deploying agentic AI for multi-step tasks that previously required teams of people. The LLM Suite is updated on a fixed eight-week cycle regardless of whether the business feels ready. That's not a pilot. That's infrastructure.
What I think most software teams are doing right now is waiting to see how serious this is. They've deployed a coding assistant for a few engineers. They've run a proof of concept. They're monitoring the results. That response made sense twelve months ago. It doesn't make sense now, when the largest financial institution in American history is publicly saying they've doubled their AI use cases in a year and have 40,000 engineers in the program.
Chris asked me yesterday if I thought the small B2B software shops were going to be okay through all this. Genuinely curious, that kind of question. I told him the ones that treat AI as infrastructure rather than a feature are going to be fine - probably better than fine. The ones that treat it as a marketing bullet point on their pricing page are going to have a bad time explaining renewals in 2026. We actually wrote about exactly that dynamic a few months back when we looked at what happens when your vendor's numbers go sideways. The pressure is real.
What I Think Is Actually Going to Happen
Here's my take, and I'm not hedging it: the JPMorgan story is the last moment where a business can credibly claim to be in "evaluation mode" on AI. That window is closed. The evaluation phase ended when a 317,000-person institution announced it has already displaced workers, is running 450 production AI use cases, and is spending nearly $20 billion to go further.
Software teams that have not made AI coding assistants standard practice are going to start losing the productivity gap argument fast. Not in theory - in actual sprint velocity comparisons, in actual headcount-to-output ratios, in actual vendor contract negotiations where buyers are asking "why do you need six engineers for this when your competitor does it with three and an AI layer?"
The Dimon acknowledgment about workforce displacement is the real headline that's being softened by everyone covering it politely. He said JPMorgan has already displaced workers. He said there will be fewer employees in five years. He said governments need to build support systems now because "it may happen faster than we can adjust to it." He's not predicting catastrophe - he's describing a process that's already underway and asking the world to prepare. When the CEO of the most profitable bank in American history talks like that, and his CFO is bragging about vibe coding in investor presentations, the timeline compression is real.
The Last Jedi gets criticized for its pacing. People say nothing happens in the middle act. But everything is happening - it's just happening slowly until it isn't. The evacuation sequence at the end is chaos precisely because no one was paying attention to the gradual attrition until it was nearly too late. I've told Tory this four times and he keeps saying "Derek that's about space ships" and then staring at his divorce papers. But I'm right about the pacing. I'm also right about this.
The mandate isn't optional. The software teams that understand that right now - not in six months when the next round of vendor announcements forces the conversation - are the ones that are going to be interesting to work with in 2027. The ones still in evaluation mode are going to have a very uncomfortable annual planning season. We've been watching the agentic AI shift move from buzzword to boardroom agenda all year. JPMorgan just made it a line item with nine zeros behind it. That's not a trend anymore. That's the ground shifting.