Intelligence Has a Meter Now
Since reasoning models arrived, how well an AI performs depends on how much thinking you pay for, and that quietly breaks the way we define, measure and sell intelligence. The opening essay of The Price of Thought.
Atlas AI Labs · An essay series · 2026 to 2036
The Price
of Thought
Intelligence is no longer just what a machine knows. It is what a system can afford to think, and how wisely it spends each thought.
By DS Praneeth
dspraneeth07
dspraneeth0701 / The question
One question, followed for a decade.
Intelligence is becoming a budget. The future belongs to the systems, and the people, who know how to spend it.
Central question
What happens to intelligence, and to everything built on it, when thinking itself becomes a metered, allocatable resource?
The argument
The defining capability of the next decade of AI will not be raw knowledge or raw reasoning. It will be allocation: knowing what is worth thinking about, how much to think, what to remember instead of recompute, and what to verify. Today's systems are strong but economically naive. They do not know the value of their own next thought. The systems, companies and engineers who close that gap will set the cost curve of intelligence itself.
Who it is for
Senior ML and infrastructure engineers, applied AI researchers, technical founders, CTOs and investors, and serious AI readers who are tired of news recaps and want a framework that still works next year.
A different lens
Most AI commentary asks “how smart is it?” This series asks “what did that intelligence cost, who decided how to spend it, and was it worth it?” It reads AI through computer science, economics and physics together.
Why it lasts
Every computing system ever built has faced the trade between computation and quality. Model names will change. The economics of thinking will not go away.
Why it matters to you
Researchers care about test-time scaling, infrastructure engineers about inference cost, founders about unit economics, executives about energy and capex. This topic speaks to all four at once.
02 / Principles
The rules every essay is written by.
01
One idea per essay
Each essay defends a single claim. If it can't be stated in one sentence, the essay isn't finished.
02
Mechanism over mood
Explain why something happens: what is scarce, what is traded, what is cheap to check. Never just say that it is happening.
03
Three-colour honesty
Every essay separates Established from Prediction from Speculation, visibly.
EstablishedPredictionSpeculation
04
Steelman, then decide
State the strongest objection better than its proponents would, then take a position anyway.
05
Old ideas, new light
Use deep roots such as Simon, Turing, Shannon, Hayek and Landauer to make new things legible. Never invent a theory and present it as fact.
06
Keep score in public
Every essay ends with dated predictions, re-scored each October in the Ledger edition. Being publicly wrong is part of being credible.
03 / The vocabulary
Ideas the series keeps coming back to.
Each concept is introduced in an essay. The rest stay sealed until they are.
- I
The Thinking Budget
Intelligence is now delivered as a quantity that can be dialled up or down. Every system has a budget, whether or not anyone designed it.
- II
The Value of the Next Thought
The central skill is to estimate whether one more unit of thinking is worth its cost, and to stop when it isn't. In the literature this is metareasoning.
- III
Opens with essay No. 03Tue 2 Feb 2027
- IV
Opens with essay No. 05Tue 15 Jun 2027
- V
Opens with essay No. 07Tue 7 Dec 2027
04 / The Ledger
Every essay ends with a bet.
Each essay closes with a dated, falsifiable ledger. Every October, from 2029 through 2036, I publish a Ledger edition that scores it: hit, miss or too early. Most commentators never revisit their predictions. This series will.
- Established
Widely reported and checkable today. Sourced in the notes.
- Prediction
A dated, falsifiable claim about the next two to four years.
- Speculation
A possibility worth watching, with no date attached.
See the ledger from No. 01
- Established
Major frontier APIs expose user-controllable reasoning budgets (effort levels or thinking-token limits), and reported results improve with test-time compute on reasoning benchmarks.
- Prediction · Oct 2028
By October 2028, most headline frontier-model launches will report cost (or compute) per task next to accuracy on at least one major reasoning benchmark.
- Prediction · 2029
By 2029, the default experience in the leading AI assistants will choose its own thinking budget per request for most traffic, with user override as the exception.
- Prediction · 2030
By 2030, a meaningful share of enterprise AI contracts will price at least some workloads per completed outcome rather than per token.
- Speculation
“Quality per joule” or a similar efficiency figure becomes a consumer-facing marketing metric for AI, the way fuel economy is for cars.
Integrity note
Every essay separates what is Established from what is a Prediction and what is Speculation, visibly, in the Prediction Ledger at its end. The notes under each essay list every source. If a claim turns out to be wrong, the correction is recorded here, not quietly edited away.
- 2029
- 2030
- 2031
- 2032
- 2033
- 2034
- 2035
- 2036
Every October, from 2029.
05 / Calendar
Ten essays, one argument, two years. Then a ledger every October.
The cadence is deliberate: tighter at the start to build a returning readership, then quarterly, so each essay has time to be shared, debated and answered before the next one arrives. The dates are public. The essays stay sealed until their day.
Opens Tue 8 Dec 2026 · 09:00 IST
41 days8 hrs40 minTitle and summary stay sealed until then.
Opens Tue 2 Feb 2027
Sealed until Tue 2 Feb 2027, 09:00 ISTOpens Tue 6 Apr 2027
Sealed until Tue 6 Apr 2027, 09:00 ISTOpens Tue 15 Jun 2027
Sealed until Tue 15 Jun 2027, 09:00 ISTOpens Tue 7 Sep 2027
Sealed until Tue 7 Sep 2027, 09:00 ISTOpens Tue 7 Dec 2027
Sealed until Tue 7 Dec 2027, 09:00 ISTOpens Tue 7 Mar 2028
Sealed until Tue 7 Mar 2028, 09:00 ISTOpens Tue 6 Jun 2028
Sealed until Tue 6 Jun 2028, 09:00 ISTOpens Tue 24 Oct 2028
Sealed until Tue 24 Oct 2028, 09:00 IST- ThenThe Ledger, every October from 2029
06 / Conduct
Rules that make a series credible for ten years.
Never miss a date
Publish on the calendar or announce the change in advance. Reliability is part of the brand.
Public corrections log
Every correction is listed on this page, dated, with what changed and why. Readers trust writers who show their errors.
Credit the best objection
Each essay names, with permission, the strongest reader critique of the last one. It turns readers into contributors.
Judged by serious readers
The measure that matters is returning readers, subscribers and replies from practitioners, not likes.
No hype, ever
No AGI dates, no “X is dead”, no undisclosed sponsorship. The series' authority depends on restraint.
Sources before claims
Every factual claim about the outside world is sourced in the essay's notes and checked before the issue goes live. Predictions are always labelled as predictions. Nothing is presented as inside knowledge of any lab.
Corrections log
No corrections so far. When one is needed it will be listed here with the date, the essay and what changed.
Send the strongest objection
Each essay names, with permission, the best critique of the one before it. If a premise is wrong, say so, and say why.
Send an objection07 / Essays
Read in order.
Intelligence Has a Meter Now
Since reasoning models arrived, how well an AI performs depends on how much thinking you pay for, and that quietly breaks the way we define, measure and sell intelligence. The opening essay of The Price of Thought.
The next essay opens Tue 8 Dec 2026.
Cite this series
DS Praneeth. The Price of Thought: an essay series. Atlas AI Labs, 2026–. https://www.dspraneeth07.in/newsroom/series/the-price-of-thought
Get each essay the day it opens.
One email per essay, with the ledger and the sources. The next essay opens Tue 8 Dec 2026.
Your address is used only to send the essays. No sharing, no tracking pixels. Privacy note.
Disagree with a premise? Good. The best objection gets named in the next essay.