---
title: "Google's own audit cannot find the AI flywheel its founder says has started"
description: "DeepMind's first internal audit of AI-driven research, published October 14, 2026, finds capability climbing a steady 0.1 index points a month with no self-driven speedup; a day later Sergey Brin said the flywheel has started. Alphabet's capital spending assumes he is right."
publisher: "The Inference"
section: "Frontier"
published: 2026-10-16T17:27:50.181Z
modified: 2026-10-16T17:27:50.181Z
canonical: https://theinference.org/article/google-s-own-audit-cannot-find-the-ai-flywheel-its-founder-says-has-started
language: en
keywords: "Google, Alphabet, DeepMind, Sergey Brin, AI Safety, Capital Expenditure"
---

# Google's own audit cannot find the AI flywheel its founder says has started

> DeepMind's first internal audit of AI-driven research, published October 14, 2026, finds capability climbing a steady 0.1 index points a month with no self-driven speedup; a day later Sergey Brin said the flywheel has started. Alphabet's capital spending assumes he is right.

Google put out two documents this week that disagree, and only one of them describes what [Alphabet](https://theinference.org/markets/companies/alphabet)'s spending is buying. On October 14, 2026, [Google DeepMind](https://theinference.org/markets/companies/google-deepmind) published [the first internal audit of whether AI is accelerating AI research](https://cryptobriefing.com/google-deepmind-no-ai-research-acceleration/), tucked into the safety report for its new frontier model Gemini 4 Argon, and found no speedup the machines could claim credit for [1][2].

A day later, Google co-founder Sergey Brin told a roundtable that his researchers and their models "work hand in hand" and that Google is "[starting to get onto that RSI treadmill](https://www.businessinsider.com/sergey-brin-google-gemini-ai-argon-recursive-self-improvement-rsi-2026-10)" [3][4]. Recursive self-improvement, or RSI, is the loop in which AI helps build the AI that replaces it, with each generation supposed to arrive faster than the last [3].

## The audit found a straight line, not a flywheel

The assessment, led by [DeepMind](https://theinference.org/markets/companies/google-deepmind) researcher Haydn Belfield, tracks frontier capability on a General Capability Index: one number per model, blended from scores across 123 models and 92 benchmarks, something like an IQ score for AI [1][2]. Since late 2024 the score has climbed about 0.1 points a month, a remarkably steady line [2]. For its public models, the monthly gain before April 2026 was 0.091 points. After April it was 0.118, and the report found no statistically significant difference between the two [1].

Argon itself sits marginally above the trend, and the team traces that to increased compute and organizational investment rather than self-acceleration [1][2]. As Belfield put it, "[There is currently no significant end-to-end acceleration of ML R&D, AI-driven or otherwise](https://decrypt.co/381061/google-deepmind-ai-isnt-building-itself-writes-most-code)" [2]. One caveat cuts both ways: the numbers describe only DeepMind's own lab, not [OpenAI](https://theinference.org/markets/companies/openai)'s or [Anthropic](https://theinference.org/markets/companies/anthropic)'s [2].

## AI writes the code, humans still steer almost every change

The flywheel would have to engage in the middle of the chain, and that is where it stalls. AI tools now write more than three quarters of DeepMind's production code, yet about 88% of code changes still need a human steering the AI or writing code before anything ships [1][2]. The share AI completes end to end moved only from 7.5% to 11.9% between April and September [2].

On RE-Bench, an internal suite of 74 machine-learning engineering tasks, the Argon agent finished about half [1]. Gemini 3.7 Flash, the previous generation, managed 27%. Both sit far below the 90% score DeepMind treats as a red flag for full automation [1][2]. Across 1,389 reviewed agent runs, the bigger the task, the less likely the model finished it alone [2].

*Chart: **On RE-Bench, Argon finishes about half; DeepMind's red flag for full automation sits at 90%** Column chart of RE-Bench completion rates: Gemini 3.7 Flash 27 percent, Gemini 4 Argon agent about 50 percent, DeepMind's automation red flag 90 percent*

*Share of RE-Bench machine-learning engineering tasks completed, per cent. The 90% column is the score DeepMind treats as a red flag for full automation, not a model result; Argon's figure is given as about half. Sources: DeepMind Gemini 4 Argon safety audit, via Crypto Briefing and Decrypt. [1][2]*

A survey of 12 senior DeepMind researchers found automation highest on well-specified, verifiable execution and lowest on open-ended ideation and long-horizon research taste [1][2]. What stays scarce is verification, review, and the judgment about which idea deserves compute.

## The spending assumes the founder's version

Brin says Google is already a couple of models ahead of Argon internally and expects the pace only to accelerate [3][4]. Vihan Jain, a Gemini RSI lead at DeepMind, posted that in post-training, the stage where a finished base model is refined on feedback and data, demand had shifted from asking for raw chips to asking for "Argon quota" [3].

The thesis moves real money. Chip stocks fell almost 6% a month ago when Anthropic's chief executive, [Dario Amodei](https://theinference.org/markets/people/dario-amodei), argued that recursive self-improvement was already accelerating [2]. A report published on Wednesday by [Gladstone AI](https://www.gladstone.ai/), a firm that advises the U.S. government on AI policy, found most researchers it interviewed expect RSI within three to twelve months [3][5][6]. Alphabet spent [about $45 billion on capital projects in the second quarter](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001652044), against a measured record in which every point of capability still comes from chips and people [7].

*Chart: **Quarterly capex has more than quintupled while the measured capability slope held flat** Line chart of Alphabet quarterly capital expenditure rising from about $8 billion in Q3 2023 to about $45 billion in Q2 2026*

*Alphabet capital expenditure by fiscal quarter, US$ billions, from SEC filings. DeepMind's capability slope is a separate index described in the text. [7]*

## Read the ledger before pricing the flywheel

None of this is a case against AI. The index climbs steadily, the agents are real, and [Argon, unveiled on September 30](https://theinference.org/article/google-prices-gemini-4-argon-at-2-per-million-tokens-a-fifth-of-openai-and-anthropic-s-10-flagships), is genuinely finishing long-horizon engineering work across Google [2][8]. It is a case for separating the measurement from the story: on DeepMind's own numbers, every point of capability so far traces back to compute and headcount rather than to machines compounding their own progress.

Two lines in DeepMind's ledger will settle the question: the monthly slope on the capability index, and the fully delegated share of code changes. A slope that steepens without new compute, and changes that stop waiting for review, would be the flywheel arriving. Until then, the value goes to whoever builds the review and trust layer between what AI writes and what an organization ships.

## Takeaway

DeepMind's audit finds capability climbing a steady 0.1 index points a month and 88% of code changes still waiting on a human; until the slope steepens without new compute, Alphabet's $45 billion quarters buy people and chips, not a flywheel.

## Sources

1. [Crypto Briefing, Google DeepMind finds no acceleration in AI research progress, 15 October 2026](https://cryptobriefing.com/google-deepmind-no-ai-research-acceleration/)
2. [Decrypt, Google Says AI Isn't Building Itself Yet—Even Though It Writes 75% of the Code, 15 October 2026](https://decrypt.co/381061/google-deepmind-ai-isnt-building-itself-writes-most-code)
3. [Business Insider, Sergey Brin says Google's AI is approaching a critical moment, 15 October 2026](https://www.businessinsider.com/sergey-brin-google-gemini-ai-argon-recursive-self-improvement-rsi-2026-10)
4. [The Times of India (via MSN), Google co-founder Sergey Brin says company's AI is helping build its own successors, 16 October 2026](https://www.msn.com/en-in/news/other/google-co-founder-sergey-brin-says-company-s-ai-is-helping-build-its-own-successors-hinting-no-slowing-down-of-ai-development/ar-AA2emUk0)
5. [TIME, A New Report Sheds Light on the Secretive Race to Build Self-Improving AI, 14 October 2026](https://time.com/article/2026/10/14/recursive-self-improvement-AI-report/)
6. [Gladstone AI, company website, retrieved 16 October 2026](https://www.gladstone.ai/)
7. [Alphabet, quarterly capital expenditure from SEC filings, retrieved 16 October 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001652044)
8. [Google DeepMind, Gemini 4 Argon model page, retrieved 16 October 2026](https://deepmind.google/models/gemini/)


---

Google's own audit cannot find the AI flywheel its founder says has started — The Inference. Canonical: https://theinference.org/article/google-s-own-audit-cannot-find-the-ai-flywheel-its-founder-says-has-started