Free cap table template for CFOs who like to plan ahead

Even the cleanest spreadsheet can hit its limits. This free cap table template by Fidelity Private Shares can help early-stage teams track equity clearly and correctly, while setting the stage for a seamless transition to a more scalable solution when needed.

Yo, it's CJ! Welcome back to my newsletter for current and aspiring CFOs. My goal is to make YOU better at your job by covering

All in a way you can actually understand.

I also help with two things that keep CFOs up at night: picking the right software and hiring the right people.

In our recent CFO Mailbag, reader "McParty" (early contender for best mailbag pseudonym of the year) asked if R&D spend as a % of revenue will ever go down.

If AI is so efficient and engineers can now conjure up their deepest darkest creations in hours rather than months, will the relative amount we spend on building new shit decrease?

At initial glance, you think duhhh that has to happen. Right? People are more efficient. So you need less people to do it. Four day work week, get the Miller Lites on ice, let's boogie.

But on my third iced coffee I realize it's a more nuanced question. Three assumptions are baked into that take, and I'm not sure any of them hold up.

A butterfly flaps it’s wings…

It costs a lot more to run LLMs than it does to buy 30 GitHub licenses. No matter how much you used to spend on Bitbucket or Trello or Linear, LLM tokens cost more. My old rule of thumb was that Engineering Labor ran about 85%-90% of the total engineering budget, with tooling picking up the other 15%-20%. That was the budget envelope’s ratio. Now it might be 50/50, and the overall pie is bigger.

And if you're lucky enough to eliminate some headcount entirely, meet your new hungry hippo: the agent that replaced them.

Tangent: It’s v. cute when someone names their golden retriever a human name, like Doug. Or Hank. Hank is always funny. But it’s not as cute when they name their agent a human name. Stop trying to make “Harold the billing agent” happen.

A $200K engineer costs you roughly $770 a day when you spread it across working days, which sounds like a lot until you realize a busy autonomous agent can burn through millions of tokens… yanking context, rewriting its own output, looping endlessly on a problem.

(At least when I lay awake at night, anxiously looping on a problem, it doesn’t show up on the P&L. This kind of thinking does.)

And if you are lucky to eliminate some headcount entirely, what you put in its place may be an agent that chews up more tokens comparatively to the $220K you paid that engineer.

So the labor costs and token costs are not hot swappable; they are not at parity. And you didn't cancel the cost center… you’re experiencing a mix shift.

Tomasz Tunguz framed it well: the new metric isn't cost per engineer, it's productive work per dollar of inference. He explained he’s getting 31 tasks a day out of his setup for $12K a year. By that math, the engineer still on the books at $100K better be producing 8x more output to justify the delta.

That's the conversation CFOs are about to start having, and a lot of engineering orgs are not going to love where it lands.

This one's interesting. On the surface, AI makes coding more accessible. And yes, I'm personally out here using Claude to build little pet projects I had no business attempting two years ago.

But there's a real difference between that and building production-ready software for a living. Like Soc2 shit. For the latter, I think the best engineers get paid more, not less. They'll credibly argue that one of them, armed with the right tools and a phat LLM budget, does the work of the full devsec, QA, backend, frontend squad from before. Companies will pony up for that. So instead of 10 engineers at 1x, you're paying one engineer at 2x. The total headcount goes down, the price per head goes up.

This is the big one. When everyone can build faster and cheaper, everyone builds more. Javon’s Paradox, or whatever.

Counterintuitively, making it cheaper to build makes it more expensive to compete.

The winners are going to take whatever they save and immediately light it on fire on the next thing. That's just how competitive markets work. And everyone else is going to be playing catch-up, which is also expensive. Either way, the dollars find their way back into development.

(Or they go back into distribution, since that is increasingly becoming a moat. Also, click here to advertise with us lol).

So here's where I land: R&D% probably dips in the short run because you're not backfilling every open req with a human anymore. But the balance between people costs and inference costs shifts, the people you keep cost more, and competition makes sure the savings don't stay savings for long.

You'll likely spend more in aggregate because competition demands it. The ante goes up.

Which, for all the CFOs at home, makes you wonder how that quest for 25% profit margins ever gets to 35% or 40%.

What if AI just competes the margin right out of our P&Ls by making all of us more efficient at the same time?

What if nirvana becomes only 15% profit margins?

What did we do!?

Now wouldn't that be a quandary. And this is why we can't have nice things.

Net net: Don't touch that R&D dial in your LT forecast just yet.

TL;DR: Medan Multiples are FLAT week over week.

The overall tech median is 3.1x (FLAT w/w).

What Great Looks Like - Top 10 Medians:

Wishing you formidable yet scalable R&D investments,

CJ