When someone asks "how’s business?", they’re really asking "how are your finances?"

Every business decision is a financial decision. Yet too many high-leverage teams spend their weeks stuck in reverse: chasing lost receipts, policing SaaS subscriptions, and grinding through a close that drags into next month.

That’s why I use Brex.

It’s an intelligent finance platform with AI workflows that absorb the manual maintenance: corporate cards, real-time policy enforcement at the swipe, and continuous reconciliation. Close takes days, not weeks.

Stop playing defense. Join Anthropic, DoorDash, and Coinbase, and take back your momentum at brex.com/metrics.

If you want a copy of the report before the release date, sign up below. It’s free. It’s comprehensive. And it almost killed me to write lol. Holy market mapz.

I recently saw Matthew Prince, CEO of Cloudflare, speak at a conference. For context, about a third of the internet's traffic passes through Cloudflare, so they have great power and responsibility in terms of keeping an open web that's accessible to everyone (including agents).

He said in 2026 traffic from agents surpassed that of humans.

Wut?

When asked about this purported inflection point a year prior, he thought we were three to five years away. Both he, and everyone else in the crowd, were blown away by how fast agents passed us mere mortal doom scrollers.

I think 2027 will be the year agentic activity within the CFO tech stack surpasses that of human activity.

Which is a pretty wild thing to surmise, since the knowledge graphs on which these agents need to travel on were not, you know, built with agents in mind.

Since we dropped our inaugural State of the Agentic Finance Stack report last year, I've recorded more than 100 podcasts with CFOs implementing new tech and interviewed dozens of founders who are building for our homies that wear both a belt and suspenders to work.

How are people feeling?

Well, there's a ton of FOMO in the air right now. It’s palpable. And a lot of it is tied to our identity as finance professionals, ringing our hands and pacing around our corner offices that someone else (or someone else's agent) will take our gig if we don't FITFO (figure it the f out).

I also think many of the developments in the space are overblown, and have yet to absorb the physics of an annual renewal cycle. Discounts are great, but the calendar is undefeated.

Speaking of that - I was speaking with a procurement leader for this report and he remarked at how the barriers to implementation, especially in the ERP and FP&A spaces, have fallen precipitously (vocab word). He's witnessed multiple customers switch ERPs not once, but twice in the same year (which probably also says a lot about that person’s choices). Multi year root canals have become more cavity fillings, coming from someone who has implemented many an erp and is eating his children’s Halloween candy as he writes this.

It’s hard to ignore all the fundraising going on in the finance tech stack. Especially with all the VC money flowing into these new GLs there's a push to try something new, and a massive pull to get the best deal.

As such, many of the finance leaders I speak to are preserving optionality when it comes to the tools they jump into bed with.

But the overwhelming ambiguity and testing is actually related to org design. Someone's gotta drive this b once it's in the garage. So how will we staff our teams and what tools will they use to get the job done?

If we can buy it is one question (once again, thanks VC funding!). If we can drive it is another.

But lucky for us, that part is getting easier too. Enter: the MCP.

In last year's report I remarked that finance hadn't had its "ChatGPT" moment yet in the same way Customer Support had theirs with a Decagon or Sierra, or Engineering with a Cursor or Codex. This year it arrived in the form of MCP. You could finally query cross-functional data using simple (and even mislpeled ) language.

Remember all those times you couldn't nail a 17 part sumif and had to resort to a series of convoluted VLOOKUPs and XLOOKUPs? Searching for customer accounts that may or may not be parent or child accounts in the state of New Mexico, but only within a certain range of zip codes, and had to have started before 2019?

But that search is entirely possible now. It’s honestly really easy.

I was remarking on the rise of MCP with Ali Hussain, CEO of Tabs.

MCP for finance, it's alchemy, but you still need to have some sort of knowledge graph for agents to travel on. It can't just be out there wandering around on an unpaved, dirt road.

That’s why I’m calling this new cohort of companies the “Class of MCP”.

MCP: It is an open standard introduced by Anthropic in late 2024.

Harness: Then you also have this thing called a harness. Think of it as the leash, the instructions, and the person walking the dog.

Headless: And one more term worth knowing: headless. Which reminds me of multiple Pusha T lines describing luxury convertibles with dat top offfff. A headless agent is one that operates without a traditional user interface.

We’re seeing UI melt away and the world is now your oyster / prompt bar. But I think agents will all be kept within protective fences that constrain or guide the direction they work in.

Well, at least they should.

A number of companies, whether that's Rippling or Ramp or now Tabs, have fundamentally understood that you need to lay the tracks for agents to travel on (employee graphs, vendor graphs, customer graphs). You have some atomic unit at the center that lets you access other data points in a way that's deterministic, and then you can layer AI on top of it.

The second order question is how we build around that.

MCP can pull a big chunk of the reporting layer, which makes the GL interesting in terms of how much AI it actually needs. The next phase is how finance presents more context to MCP, and I think that's what people will be really hungry for over the next 12 to 18 months.

And these new ways to access data for insights permeate through all layers of the finance tech stack.

Which we’ll explore AT LENGTH in this report!!!

In this report we cover new and existing players across the following layers of the tech stack:

My goal is to make you more educated buyers of technology. That required me to interview more than 30 people and individually research 145 companies across the 12 layers we cover.

A big shout out to the following partners and thought leaders who made this report economically viable to spend months researching:

I should probably go get some sunlight now. After +120 days of working on this, my Lenovo skin tan is turning green. But in the meantime, sign up to get the report early! It’s dope.

The last two years have been the greatest subsidy in history.

Investors have plowed trillions of dollars into AI that is now only translating itself into the pricing and cost for the rest of the economy.

The last time I remember this happening (albeit at a smaller level) was the ridesharing wars of the 2010s. Uber and Lyft became widely available in Boston during the winter of my senior year of college. For broke college kids trying to get to a bar in Brighton, it was a revelation that we could now call a ride from our phones for 1/3 of the price of a cab, and not walk to a cab stand in the freezing cold.

The same thing is happening with inference costs. If you get a bill for Anthropic for $60K, I bet you it’s actually closer to $150K to $200K in actual services you’re getting.

So who’s going to foot the bill? Will it get passed on?

Back to my Uber example, flash forward three years, I’m a working professional, and it’s now become cheaper to take the yellow cab when I land at Logan Airport (not to mention, they had moved the rideshare pickup approximately 738 miles away, but I digress). Uber had to flip its business model to something more sustainable, and was now passing the costs on to its customers.

History repeats itself, and it's about to get real. The way we’ve bought software in the past will evolve. And a larger share of vendor COGS will get passed through to end customers. If you're a buyer of software (hint: we all are) you should prepare your P&L to get a bit heavier.

That’s what me and Ali Hussain of Tabs discuss on this special episode of Run the Numbers.

Wishing you make good use of your definitions so people understand your business,

CJ