AI Unit Economics · Interactive Companion

The six levers that move AI cost.

AI vendors charge for every word the AI reads and every word it writes back. The path you choose to do the work changes the bill more than the model you pick. This page lets you move each lever and watch cost respond — same job, twenty times the cost depending on the design.

Start here

Three ways to build the same thing

What happens when you change one variable

Pricing source view source ↗

Model choice Sonnet 4.6
Capability tier sets the per-token rate. Output is typically priced 5× input.
Pages you give the AI per round 15 pages (≈ 16,500 tokens)
Documents, briefs, or context you send. Re-read on every round.
Length of the AI's answer 600 words (≈ 800 tokens)
How much the AI writes back. Words cost ~5× more than the same volume read. Reasoning models add hidden thinking tokens you still pay for.
Model style:
Back-and-forth rounds 4
How many times the AI works through the problem before answering. Each round re-reads everything from earlier rounds.
Pages each lookup returns 2 pages (≈ 2,200 tokens)
When the AI searches a database or document, how much it pulls back. Stays in the conversation from then on.
AI specialists working in parallel 1
One generalist, or a team — researcher, analyst, writer. Each specialist runs the full conversation independently.
Times this runs per year 4,000
How often this task runs. Drives the annual budget, not per-task cost.
Cost per run
$0.85
vs simplest version: 17×
Total reading by the AI 110 pages
Total writing by the AI 1,800 words
Cost of reading $0.34
Cost of writing $0.51
Cost per run $0.85
× times per year 4,000
Annual envelope $3,400

Each lever explained

1 · Model choice — the easiest lever, the one teams ignore

Cheap to premium models can be a 15× rate jump. Most production work doesn't need the top tier; a triage-then-validate pattern (a fast cheap model filters, a stronger model adjudicates the shortlist) captures most of the quality at a fraction of the cost.

Match the model to the decision, not the workflow.

2 · How much you give the AI — the silent compounder

Every page you send is re-read on every round. A 30-page protocol re-read four times in a back-and-forth workflow is 120 pages of reading for one job — before the AI has written a single word back. Documents with images and tables cost 2–3× more than plain text.

Send only what's needed. Caching helps when the documents don't change.

3 · Length of the answer — costs 5× more than reading

What the AI writes back is the most expensive part of the bill. Verbose responses, long reasoning explanations, and structured-output overhead all bill at the premium rate. "Reasoning" models also produce hidden thinking-text you still pay for.

Constrain answer length. Ask for what you actually need.

4 · Back-and-forth rounds — the non-linearity

A 4-round workflow doesn't cost 4× a single call. It costs roughly 10× because round 4 re-reads the entire conversation history from rounds 1 through 3. This is where naive forecasts break, and where most of the hidden cost lives.

Cap rounds in code. Set ceilings on agent loops.

5 · Lookups — every search result rides along

When the AI looks something up and the database returns 2 pages of results, those pages enter the conversation and get re-read on every round after that. Five lookups in an 8-round workflow can quietly add 40 pages of reading to the bill.

Summarize lookup results before re-reading. Drop stale findings.

6 · Specialists in parallel — N independent conversations

Multi-specialist designs look elegant in a diagram. In billing terms, each specialist runs its own full conversation. Three specialists on a 4-round workflow cost roughly 3× a single specialist's full run — even if they're collaborating on the same job.

Parallelism multiplies. Sequential delegation can be cheaper.

7 · Model verbosity — the lever this tool keeps qualitative

Different models produce different output volumes for the same task, independent of the rate card. A more expensive model can be more concise and still cost more, because the rate jump exceeds the token savings. A "reasoning" model can produce 3–5× the visible output because the hidden thinking is also billed. The toggle above lets you explore the directional effect.

Cross-model comparisons here are directional, not like-for-like.