Some companies are rebuilding themselves around AI. Everyone else is waiting for a lab, vendor, owner, or competitor to do it for them.

Token Economy Part 1 said tokens don't create productivity. The operating model does.

This week shows what happens next: if you can't build that operating model yourself, someone will install it for you.

Sometimes the universe is holding out a giant flashing sign saying pay attention to this pattern. 

Companies with intelligence in their operating model are shipping so much faster.

And the companies that ship fastest are also growing their revenue fastest. There’s a direct correlation. This is happening against the backdrop of companies like Coinbase announcing 14% of staff will be let go, Atlassian 10%, and with more to come. 

Waving away AI as hype is dangerous because there are levels to AI adoption. There are levels of intelligence for your operating model. And if you’re not higher up that curve, you’re not feeling the benefit in productivity or outcomes.

Most companies are stuck at level 1. A few outliers like Ramp and Anthropic are at level 4. Nobody is at 5 yet.

Who installs this increasing intelligence? You, or someone else?

Tokenmaxxing argued that the winning company converts tokens into shipped outcomes. 

Previously, I outlined 4 layers of competitive advantage for companies building their operating model around intelligence, Talent, Data, Model an Agentic Workflows. Each layer compounds the one below it. 

The deployment question is: which layers can you build yourself, and which require outside help? Take the FIS example, they have the data, but did they have the talent to take that, and either build their own model or agentic workflows? Arguably, not at first. So they brought in Anthropic.

If your operating model is the product, then whoever rebuilds your operating model owns a piece of your moat. 

Transform or be transformed (one shall stand, one shall fall Megatron).

There are three clear paths emerging to having intelligence transform your operating model. 

There’s also secret option 4, get left behind. You read all of this. You agree it's important. You'll do something about it next quarter. A year passes and Copilot usage is up but your operating model hasn't changed. 

They differ by which layer they choose to compete on, and by who sits in the room while the work gets done.

Ramp gives us the clearest public picture of Level 4. A company-wide operating system: shared skills, persistent memory, internal tools, token observability, and non-engineers building useful software. In Ramp cracked enterprise AI adoption, I wrote about the custom harness every employee now has. 

You can see the results in the metrics Geoff Charles shared in how to get everyone in the company AI-pilled: 

The difference between Ramp and every other company that "set an expectation around AI": Ramp built the ladder. Expectation without infrastructure is just an email from the CEO.

What I love about the Allica example is how similar some of these parts are. They talk about connecting every internal tool via MCP (Jira, Confluence, Figma, Hubspot), running internal spotlights, do skills sharing, and they’re fundamentally changing their org design around the new workflows that have emerged. Roles are getting smaller, consolidating frontend, backend, and QA into full-stack engineering, and teams are getting smaller.

Every time I speak to one of the AI labs, they tell me financial services is their hottest enterprise segment. The enterprise GTM and solutions teams are as shocked as I was about that, but whether it’s coding agents, enterprise licenses, or partnerships with the labs, the incumbents are not behind on spending money. Where they are behind is in the level of adoption. Where Allica is Level 3, Ramp is Level 4, even the most tech-forward incumbents are at Level 2 (meaning, they haven’t yet altered how the entire company operates, just pockets). 

One way to shortcut that is to partner with the labs solution and forward-deployed engineering (FDE) teams, and just last week, Anthropic announced its partnership with FIS. 

If you’re unfamiliar, FIS is one of the largest systems of record for banks, powering ~12% of the global economy.” They do the core systems, payment rails, and deposit and lending systems of record. 

As I wrote a few weeks ago: 

The trade is interesting on both sides. FIS gets an AI story without cannibalizing its core overnight. Anthropic gets distribution into one of the most regulated, workflow-heavy, high-value sectors in the world without banging on procurement doors at 2,000 individual banks. Banks get something that looks less like a moonshot and more like an upgrade path from one of their existing critical suppliers (who the regulator is comfortable with too).

Strategically, this is a very smart move for Anthropic. The labs need distribution. Distribution into regulated industries is the hardest distribution to get. Partnering with the system of record bypasses two years of enterprise sales cycles. The FDE-led, knowledge-transfer model means each engagement teaches the lab something it didn't know about how its tools survive contact with real workflows.

Not to be outdone, Customers Bank is a $26 billion-asset Pennsylvania-based commercial bank known for its technology-forward approach. In April 2026, they announced a multiyear partnership with OpenAI to embed OpenAI engineers on-site and co-develop AI-native tools. Their goal is to cut the banks cost to income (CI) ratio from 49% to 40%. OpenAI gets a reference client, and Customers Bank gets a head start.

But I have several questions about this model, like. What does an FIS or Customers Bank have to agree to with the AI lab in this case? I’d imagine there’s some term limit of exclusivity, but also, in time, they’d work with every lab. What happens to the partner ecosystem around the system of record who ran AML, credit, onboarding, and fraud? Those specialists are trying to move up the food chain into agentic workflows themselves, but don’t have the data moat that an FIS has. 

The practical reality here is, the base models keep improving, but the specialists benefit from those too. Take Hebbia AI, a company working in financial markets workflows. Anthropic has released dedicated financial services agents for workflows like KYC and Treasury. And while they’re good, they’re not specialist good. 

As a metaphor, consider that Claude Code and OpenAI Codex coding agents haven’t slowed down Cursor one bit. But they have created a massive amount of enterprise revenue for the labs.

With all of that said, in this model, FIS / Customers Bank is playing to its strengths, and keeps its moats and autonomy for the most part. What’s unclear is if they can ever internalize that skill or end up relying on the labs.

The third mode is the one that should make every private equity portfolio company CEO sit up.

Private equity has always used levers like pricing discipline, cost takeout, and shared services as its way to transform the economics of a company they buy into, and ready it for sale. AI is a new lever, and the $1.5B Anthropic-Blackstone JV is the explicit operationalization of it. The owner writes the cheque, and Anthropic engineers show up at the portfolio company. Map the workflows. Identify the highest-leverage operational changes. Ship. And then Benchmark.

As a leadership team, you’re now benchmarked against another portfolio company that runs customer support with 40% fewer people after the engagement. Underwriting cycle times, shipping velocity, PR’s merged all become benchmarks. It’s never apples to apples, but its a point of pressure.

And leadership teams might not get a long time to implement the turnaround. If they can’t transform themselves, and the FDE’s are doing more of the work, what is the leadership team still doing there? And if you send in the FDE’s and they still don’t perform, should the PE firm exit that company even at a loss?

As an aside: There’s a real lack of neutral FDE talent (outside of Palantir and a handful of companies starting up to do exactly that). Take FDEs from a lab, and you get exclusivity tilted toward their model family. Take FDEs from your owner, and you didn't have a choice. A neutral FDE market exists but is thin. That gap will close in 18 months. Until it does, "let the lab inside" carries a time-bound model-family bet most CIOs haven't fully priced, because everything in AI changes faster by the day.

The uncomfortable question for the boardroom is: if the FDEs ship more in 90 days than the internal team shipped in a year, what exactly is the internal team's job?

And if the answer is "context" — knowing the customer, the regulation, the edge cases — that's a real answer. But it's a shrinking one. Every engagement teaches the lab more about those edge cases. 

Fintech that does nothing with AI is dead. 

The sheer fact that giant banks like RBC have started to build their own foundation models, and like BBVA have entire teams whose workflows are now re-oriented around AI says their operating model is changing.

But the companies I worry about are the ones that think having 99% adoption is enough, where everyone has Co-pilot, and the engineers are all using Cursor, but nothing in their operating model has changed from a year ago, and the tools and rhythms of the company look exactly the same. 

This applies as much to Fintech companies as it does to banks. You’re all incumbents now unless you’re moving uncomfortably fast up through the levels of AI operating model capability towards intelligence driving everything in the company.

The companies that don’t move up have two endings. One: acquisition or restructuring by an owner who'll do the transformation for them, on the owner's timeline, with the owner's chosen labs, with the owner's people in the room. Two: they don't.

Org charts can get smaller. Jobs can get bigger. And there’s a theme emerging of companies doing exactly that. Some are doing all three.

Block cut 40% of headcount, limiting levels between CEO and every employee to 5, and changing the roles of PMs, designers, and engineers. Coinbase and Atlassian have both made similar announcements. Alicia and Ramp haven’t announced job cuts, but again, teams are smaller, each person does more, and in aggregate, the company’s output increases.

There’s some abstract principles that apply to everyone. 

The companies that brought in the FDE’s have not yet changed their operating model. They get a new AI feature, without becoming the type of company that can build those features. 

Teams that have made this shift are publishing blogs about how much cycle time fell, or which cost line moved, or how much quality improved. The teams that have hired FDE’s have put out press releases saying they’ve hired FDE’s. And that’s not to throw shade at them, for them, that is the smart move. But only if it’s backed up with long-term results. 

Companies rebuilding their operating model around intelligence benefit from the best structural tailwind the economy has ever seen. Token prices keep falling by roughly 10x every 18 months. The outputs those tokens produce keep getting more capable. Models are swallowing entire tasks, then entire industries.

Cheaper intelligence lowers the cost of installing entirely new operating models inside companies that haven't built one themselves. And if your operating model assumes tokens will get cheaper and better, you don't have to retrain your workforce. 

Those things just happen as the labs update the models.

The labs are now in the operating-model installation business. 

Private equity is funding it. 

The companies that built their own — Allica, Revolut, Ramp — are pulling away. 

The ones that didn't are about to find out who owns them.

ST.

1. Shift Markets - Prediction Markets as a Service

Shift Markets provides a white-label prediction market capability that any brokerage or platform can embed into their offering. It connects to liquidity at both Kalshi and Polymarket and has helped 25+ brokerages launch to more than 10m users. 

🧠 The company started doing crypto as a service but is now leading with prediction markets. Like just about every major brand in crypto rn. It’s a solid idea, I imagine a lot of companies want to embed all prediction markets, not just partner with one, but it makes you wonder. Are Poly and Kalshi consumer brands first, or do they increasingly become Nasdaq-like infrastructure?

2. Pmtbox - E-commerce orchestration + revenue optimization

Pmtbox unifies disparate PSPs, fraud providers, data tools and customer service layers to improve conversion, give insights into ways to grow revenue and stripe out fees and costs that are not needed. It’s used by companies like Domino’s and the Salvation Army.

🧠 Weird name, surprisingly powerful product. The out-of-the-box fraud tool from your PSP is rarely well optimized; some fraud vendors also block a lot of good transactions that could have been a sale. Managing that via multiple dashboards and APIs is a pain. Pmtbox fixes that. But my goodness, get a better name guys. PMT? Seriously.

3. Shiprazor - eCommerce shipping logistics for South Africa

Shiprazor integrates up to 20 different courier providers to ensure merchants always have the best option available for customers at checkout. Customers see on average a 25% increase (!!) in conversion, and the service can also handle returns and customer support.

🧠 In many markets, we’re now spoiled for courier options in e-commerce, but that’s not true everywhere. Amazon found conversion increases every time they sped up delivery. Bringing that globally is a huge opportunity. The TAM of people without good logistics is enormous.

4. Chord - AI Agents for Commerce Ops

Chord helps merchants deploy agents across their existing commerce stack to identify new potential customers, find revenue opportunities, and optimize spend. Already used by customers like SONOS, Blue Bottle Coffee, and Mr. Beast.

🧠A lot of tools like Shopify or Stripe have these agent tools baked in, but most merchants don’t have a clean, single PSP or provider stack. So having agents that can operate across disparate systems, and answer questions in natural language helps solve for that bigger world view. 

1. Senate Banking Committee Advances Clarity Act

The Digital Asset Market Clarity Act of 2025 (CLARITY) advanced out of committee with a 15-9 vote with two Democrats joining with Republicans making it bipartisan. The vote follows months of debate and discussion over “yield” in Section 404, as well as key sticking points like developer and tech neutrality.

🧠The compromise on yield appears to have held for the banking committee. But much can change between now and any vote being passed on the Senate floor.

🧠The proposed compromise matches credit card rewards. Banks fear stablecoin rewards become uninsured deposit competition. Crypto firms argue rewards are commerce incentives, not bank interest. The compromise appears to ban passive balance-based yield while allowing bona fide activity or transaction-based rewards. For example, credit cards offer “cash back” but do not offer yield tied to balances. The lack of clarity in CLARITY is clearly a sticking point.

🧠Polymarket is now pricing a 60% chance of the bill passing this year. This half of the bill now needs to be combined with the Senate Agriculture Committee’s half to complete a full vote. With summer recess coming and then midterms it is still a thin margin for error.

🧠 While the debate is about yield this bill does SO much more. It brings all digital assets into a clear market structure regime. Stablecoins / GENIUS are to cell phones what market structure / CLARITY is to cell towers/cell networks. You have the breakthrough technology (stables), but it’s useless unless the rails/distribution are built out / can proliferate. CLARITY provides the network-layer / baselayer protections that allow for neutral settlement substrates, as well as the regulatory definitions required for capital investment in that buildout.

🧠 The real prize is programmable capital markets. If this passes, crypto stops being just a parallel casino and starts becoming a legal upgrade path for securities, collateral, settlement, and liquidity. BlackRock and Fidelity can move with wrappers. Banks need rules before they move their balance sheet. CLARITY is the rulebook they have been waiting for.

🧠 The US already has the deepest capital markets in the world. If it gets digital asset market structure right, those markets can become 24/7, composable, and software native. That is much bigger than whether a wallet can pay 4% on a stablecoin balance. This tweet from Citrini nailed it: “Morgan Stanley gets its price discovery from Hyperliquid TradeXYZ.”

🧠 Does anyone else love that the yield is in Section 404? Yield not found. 😀

2. Anthropic launches small business agents

Just a week after launching agents for financial services and partnering with FIS to deploy new KYC agents, Anthropic announced: “Claude for Small Business is a toggle install that puts Claude to work inside the tools small business owners already use: Intuit Quickbooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365. From these tools, it can plan payroll, close the month, run a sales campaign, chase invoices, and more.”

🧠 Anthropic has streamlined workflows that are an entire category of Fintech companies. Use cases like “automate month-end closing” or view a dashboard of pipeline, sales, and trend movements are entire companies that have filled up my “4 Fintech companies” category for the past year.

🧠 The threading together of Canva, Hubspot, PayPal, and QuickBooks is a signal of the strategy. Nobody has done this as an operating system. There are AI platforms for marketing campaigns, or CFOs, but having it all in one place from Anthropic cowork in many ways is the value here. Buy one product, and it plugs into everything you do with pre-built workflows.

🧠 And late in the week Fiserv announced “agentOS” in partnership with OpenAI. “agentOS will initially feature four Fiserv agents: Commercial Loan Onboarding, Daily Operational Analysis and Reporting, Agentic Deposit Intelligence, and Agentic AML Triage Analysis.” - Is this just a marketplace for agents that banks might never use? Or do these agents work and add value? They’ve not published results so we don’t know.

🧠 Everyone wants the AI story, but the labs are it.

1. Alica Bank fully automated SMB lending with AI

Alica bank is one of the fastest growing companies in finance in Europe, one of the fastest in the UK, period. And in the past year they 3.8x’d the PR’s they’re merging with AI, and increased page build time by 89% with their AI x design system. The thing that caught my eye though, is how they’ve managed to automate SMB lending. Everything from the customer emailing, document collection, and decision-making with no human in the loop. They got to 50% across all lending, and they do complex, edge-case-filled lending. There are fintech companies who do parts of this. But a bank, built by itself, in production? 

2. Where the goblins came from

From GPT 5.1 onwards, the models would increasingly reference goblins in many harmless, sometimes quite sweet ways. But the behavior was consistent. This quirk only amplified in Codex 5.5 where, they were still funny, but more consistent. It turns out it came from a reward feedback loop during training, for the nerdy personality type. A playful style was rewarded, a verbal tic kept getting rewarded, it appeared more often, and then got specifically fine-tuned into the model. 

🧠 This seemingly harmless quirk, shows how powerful rewards in training can be. And how those rewards get amplified over model generations. What if we accidentally reward the wrong thing?

3. 📺The Karpathy Interview with Sequoia 

The guy who invented the term “vibe coding” made self-driving work for Tesla, and co-founder of OpenAI drops a ton of useful insights about how his workflows have changed since the agentic coding explosion in December 2025. I had Gemini summarize it here. 

4. 📺 Sam Altman and Patrick Collison: Possibly the best interview I’ve seen Sam give.

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out