Pricing research for operators.
Field notes on AI tooling, cloud platforms, and developer SaaS costs — written around concrete billing mechanics instead of vendor headlines.
The Real Cost of AI Coding Assistants in 2026
Cursor, GitHub Copilot, Claude Code, and others: we ran the numbers across 200+ developer workflows. The results are higher than most teams expect.
Built for budget planning, vendor comparison, and technical due diligence.
Vercel vs AWS vs Railway: True 12-Month Costs for a Real SaaS Product
We built and ran the same production application on three platforms for a full year. The all-in cost difference might surprise you.
Hidden Costs in Popular Developer Tools (Most Teams Miss These)
The fees, seat minimums, overages, and gotchas that quietly turn a $99/month tool into a $400+/month reality.
The True Cost of Adding AI Features to Your Product in 2026
Everyone wants to ship AI. Almost nobody models what it actually costs at scale. Here is the real unit economics.
Scrollytelling deep-dives
Animated, source-backed essays on the mechanics behind AI and cloud bills — each one a single idea told through a live chart you scroll.
Anatomy of an AI invoice
Everyone budgets for inference. Then the bill arrives and the model tokens are the small slice. Watch a single AI product mature from prototype to scale — and watch the dominant cost quietly migrate from the GPU to the humans checking its work.
How a $99 tool becomes a $420 tool
The number on the pricing page is an opening bid. Seat minimums, usage overages, and the enterprise-feature tax do the rest. Watch the invoice assemble itself.
LLMflation
The cost of a token has fallen perhaps a thousand-fold in three years — one of the steepest price declines in computing history. So why is the AI line on every budget going up? Two curves, pointing opposite ways.
Rent, reserve, or own
The cheapest way to buy a GPU-hour depends entirely on one number: how much of the time the GPU is actually working. Hold five options on a cost-versus-flexibility map and raise utilisation from idle to flat-out — the winner moves clean across the board.
The 15× Multiplier
Hand a hard research question to a team of agents instead of one and the answer gets ninety per cent better. The bill gets fifteen times bigger. The whole discipline is learning when that is a bargain.
The 2.7× you didn’t budget for
An AI feature’s sticker price is the cheapest number you’ll ever see it post. Follow one feature from prototype to scale — and watch the fully-loaded cost pull away from the per-token quote.
The 25 Per Cent Bill
Finance approves the build. The build is barely a quarter of what a production agent costs over three years. The rest hides in the run-and-maintain horizon — and it is where most AI business cases quietly fall apart.
The great model price collapse
Frontier inference is getting roughly 10× cheaper every year. Scroll the last two years and watch the per-token price fall off a cliff — and watch who keeps winning the race to the bottom.
The margin bridge
Take
The observability bill
Monitoring starts as a rounding error and ends as one of your largest infra line items. Scroll your fleet from ten hosts to seven thousand and watch the composition of the bill quietly flip — from the metrics you chose to the logs nobody throttled.
The Optimization Playbook
There are six levers that reliably cut an agent’s bill, and they are not equally powerful. Pull them in the right order — the free ones first — and a runaway workload becomes a budgeted one. Pull the wrong one and you spend more while believing you economised.
The Quadratic Trap
A ReAct agent re-sends everything it has ever seen on every call. That one fact turns a ten-step task into a forty-three-fold bill — and makes the tidy per-call estimate on your slide a work of fiction.
The RAG tax
“Just add retrieval” sounds free. In practice, grounding a model in your own data wraps the inference quote in a stack of supporting costs. Scroll to assemble the real monthly bill.
The real cost of AI coding assistants
A