The ROI Ledger: What My $200 Claude Code Subscription Actually Buys

A few friends have told me, gently, that I am wasting money. Two hundred dollars a month on a Claude Code subscription, every single month, just so I can call myself an AI person. They are not wrong about the cost. They are just looking at one column of the page.

This is the other column.

The reframe: a line item, not a gadget

I started this subscription in February 2026. Around the same time I made a small decision that changed how I think about it. I stopped treating it like a gadget I had bought and started treating it like a line item in the cost of running my life.

Rent is a line item. Utilities are a line item. A gym membership, an accountant, the insurance I will probably never claim. Each one is a recurring cost that I justify by the value it returns, not by the size of the number. Nobody looks at their electricity bill in isolation and declares it a waste. They look at what the electricity does.

So I gave the subscription the same treatment. Every month it costs me a fixed amount. Against that, I keep a ledger of what it actually produces. In a year, maybe two, I will sit down and do the real maths. Total paid on one side, total value created on the other. If the value clears the cost by even a single cent, after I have been honest about the discounting, it was a good deal.

That is the whole model. The rest of this post is the ledger, the maths, and the one entry that quietly outvalues all the others.

The shape of the ledger

The value side is not one number. It comes in four distinct kinds of entries, and they are worth separating because they answer different objections.

graph TB
    Sub["$200 / month
Claude Code subscription"]
    Sub --> C1["1. Things I would
never have done
net-new outcomes"]
    Sub --> C2["2. Work I would have
paid someone to do
time bought back"]
    Sub --> C3["3. Income I helped
myself earn
partial attribution"]
    Sub --> C4["4. Services I quietly
replaced
substituted cost"]
    style Sub fill:#FF5A4E,stroke:#FF5A4E,color:#0B0F14

Let me walk each one, with a real entry from my own ledger.

1. Things I would never have done at all

This is the category the skeptics never price, because it has no obvious before-and-after. It is not “the same thing, cheaper.” It is “a thing that did not exist.”

I always wanted a YouTube channel. For years it stayed on the someday list, because the someday list is where ambitions go when the activation energy is too high. Research the topic, write it, record it, edit it, design a thumbnail, publish, repeat. Any one of those is a hobby. All of them together is a second job.

The channel exists now. The Agentic Engineer. And it runs on a pipeline that researches the topic, produces the video, and publishes it, mostly without me sitting in the chair for every step. The honest point here is not that editing got cheaper. The point is that the channel exists at all, and it would not have, because the friction that kept it on the someday list is exactly the friction that an agent absorbs.

You cannot put a clean dollar figure on “a thing I wanted for years now exists.” But you cannot leave it off the ledger either. That would be the bigger accounting error.

2. Work I would have paid someone else to do

This one is closer to a normal expense calculation. There is a market price for the work, and the subscription does the work instead.

Video editing sits here. Topic research sits here. So does a surprising amount of life admin that has nothing to do with the brand. The last family vacation I planned used to be an evening lost to twenty open tabs, comparing routes and timings and prices and visa rules. That evening is back in my pocket now.

When a task has a freelancer rate attached to it, the maths is almost too easy. Hold that thought, because we will put real numbers on it further down.

3. Income I helped myself earn

This is the category people argue with most, so let me frame it carefully and keep it illustrative rather than personal.

Imagine someone deep in a job hunt. They use the tool to pressure-test their resume against each role, to rehearse interviews across very different company types, to think through compensation strategy and the actual words to use in a negotiation. Suppose they land the role with a twenty percent raise.

Did the tool earn that raise? No. They did. But did some slice of that outcome trace back to the preparation? Almost certainly yes. And here is the part that breaks the skeptic’s arithmetic: a raise is not a one-month event. It compounds across every paycheck for years. Net even a small fraction of a single year’s increment against a few months of subscription, and the comparison stops being a comparison. It becomes a rounding error in your favour.

You do not need to win this category by a lot. You need to win it once.

4. Services I quietly replaced

The last category is substitution. Things I used to pay for, or would have, that now have a different supplier.

I run a fair amount of my health through what I have started calling my personal health officer. My cycling, the muscle pain that comes with it, my sugar intake, the trade-offs in my diet. This is the kind of ongoing guidance I would otherwise have paid a nutritionist or a coach a monthly retainer for. It does not replace a doctor, and I am careful about that line. But for the day-to-day “should I be doing this differently,” it is always available and it remembers the whole history of the conversation.

And then there is the entry I value most and can price least.

I built my father a Hindi-speaking assistant he can talk to about his medicines and his health.

He is a senior citizen who is not comfortable with technology and not comfortable in English. Now he has something patient, always awake, and fluent in his own language, that he can ask about his medicines without feeling like a burden to anyone. Call that priceless and I will not argue with you. But notice that it is also a real, shipped thing that simply did not exist before this subscription pushed me to build agentically. Priceless and tangible at the same time. That is allowed.

The honest accounting

Here is where I try to be harder on myself than my critics are, because a ledger you are not willing to audit is just a feeling with a spreadsheet around it.

On the cost side, it is not only the sticker price. It is the monthly fee multiplied by the months, plus the cost of capital. That money could have sat in an index fund and compounded. Pretending otherwise would be cheating my own maths. So I will discount the value side to account for it.

On the value side, I price everything conservatively, and I leave the unpriceable entries (the channel existing, my father’s assistant) noted but not counted, so the dollar total is defensible even to someone who refuses to believe me.

To make the model concrete, here is a worked example. These are illustrative figures, not my real spend. They exist to show the shape of the calculation, not to report my bank statement. The actual numbers are what I will publish in the follow-up, a year from now.

Ledger entry (illustrative)	Conservative monthly value
Video editing I would have outsourced	$150
Research and admin time bought back	$200
Health guidance vs a coach retainer	$120
Income attribution (amortised, heavily discounted)	$300
Counted subtotal	$770
Net-new outcomes (channel, father’s assistant)	noted, not counted
Monthly cost	$200

graph LR
    Cost["Cost column
$200/mo
+ cost of capital"]
    Value["Value column
conservatively priced
+ the uncounted entries"]
    Cost -. "the only column
the skeptics read" .-> Verdict
    Value -. "the column that
decides the question" .-> Verdict["The 1-cent rule:
clear cost by $0.01
and it was worth it"]
    style Value fill:#FF5A4E,stroke:#FF5A4E,color:#0B0F14
    style Verdict fill:#F5F0E8,stroke:#5A544B,color:#0B0F14

Even with deliberately cautious numbers and nothing counted from the category I care about most, the value column is not close to the cost column. And the decision rule was never “win by a mile.” It was one cent. The skeptics are not wrong that two hundred dollars is real money. They are just reading a page with the right-hand column torn off.

The return that never reaches the invoice

I could stop here and the post would make its point. But I would be hiding the most important entry, and it is the one that is not on the ledger at all.

Because I have these tokens sitting there, paid for, waiting, I have changed how I work. I push to do everything agentically, because leaving the capacity idle feels like leaving the lights on in an empty house. I run workflows remotely so I am not babysitting a prompt from my laptop. I set up loops that get work done while I am asleep or on a flight. I have become slightly obsessed with getting more out of every token.

And chasing that, almost as a side effect, I keep playing with the actual craft. Harness engineering. Loop engineering. Skills. Subagents. How to keep context lean. How to trade latency for cost and back again. None of this was the goal. The goal was to justify a subscription. The byproduct is that I am becoming an agentic engineer.

Even if the dollar ledger came out exactly flat, the skill I am compounding as a byproduct is the asset that actually appreciates.

That skill makes the rest of my life more efficient, which feeds back into the ledger above. It also opens doors I have not walked through yet, in a job market that is starting to pay a real premium for people who can actually build this way rather than just talk about it. The two hundred dollars buys the tools. The habit those tools force is the thing that pays, and it keeps paying long after any single month’s invoice has cleared.

This is the part the “you are wasting two hundred dollars” crowd genuinely cannot see, because it does not show up as a line item anywhere. It shows up as who you are becoming.

Part 1 of a ledger I am keeping in public

So this is the deal I have made with myself. I keep the ledger honestly. I discount for cost of capital. I leave the priceless things uncounted so the maths stays defensible. And in a year, maybe two, I publish the real numbers and let you decide whether the skeptics or I had the better arithmetic.

Consider this Part 1. The follow-up will have the actual spend and the actual return, no illustrative figures, no hedging.

In the meantime, I am genuinely curious about you. Do you keep a ledger like this on your own AI tools, or are you only reading the cost column?