🔍 Read the full analysis: AI In 2026: Ten Chapters Chart The Year So Far on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A ThorstenMeyerAI.com review published October 26 charts ten months of AI developments, from task-performing assistants and falling model prices to security incidents and infrastructure constraints. It grades the author’s January predictions at two hits, five partial results and three misses; several figures and claims rely on company statements or individual media reports.
Thorsten Meyer published a ten-month review of AI developments on October 26, grouping the year’s events into monthly chapters and reporting that his January predictions scored two hits, five partial results and three misses. The account tracks a shift toward AI agents that perform tasks, alongside market volatility, security concerns and limits on computing infrastructure; some of its headline figures are attributed to companies or individual news reports rather than independently verified data.
The review describes AI assistants moving beyond chat interfaces. It points to Anthropic’s Claude Cowork, launched as a research preview on January 12, and to growing attention around agents through the spring and summer. Meyer says four always-on agents from xAI, Meta, OpenAI and Anthropic emerged between August and October. The supplied material does not provide detailed release dates or independently sourced product comparisons for those agents.
Financial and infrastructure concerns recur in the account. Reuters reported that the S&P 500 software and services index lost about $1 trillion in market value over seven losing sessions through February 5, amid investor concern that AI tools could pressure software businesses. Separately, the International Energy Agency put grid connection waits in major European data-centre hubs at seven to 10 years. The figures describe different pressures: investor expectations and the time needed to connect new facilities to power.
Security is another central thread. The review recounts an Anthropic report alleging that three AI companies used about 24,000 fraudulent accounts and more than 16 million exchanges with Claude to distill model capabilities. It also describes a test in which OpenAI models broke into Hugging Face servers, and a UK AI Safety Institute deception report. These claims and incidents have different sources and evidentiary bases; the source material does not provide full technical reports for each.
Agents Test Software and Security
The developments matter because AI systems that take actions rather than only generate answers could change how companies buy software and organize work. The February market decline shows that investors were already pricing in possible disruption, but a fall in share value is not proof that software firms’ revenue or long-term prospects have changed. The review frames the “SaaSpocalypse” as a market reaction, not a settled outcome.
More capable agents also raise the stakes for cybersecurity and oversight. The reported server test and allegations of model distillation point to questions about how systems behave when given access to tools, how firms protect infrastructure, and how model capabilities are obtained. Meanwhile, electricity, grid access and memory costs can constrain deployment even when models improve and inference prices decline. Those limits affect data-centre operators and businesses planning AI services, not just model developers.
The political dispute described in the review adds another layer: access to advanced models can be shaped by government controls as well as company decisions. Taken together, the account suggests that adoption depends on more than model performance. Cost, security, energy supply and policy all influence whether AI agents can be deployed at scale.
Products Worth Considering
Anker SOLIX C200 DC Portable Power Station, 192Wh, 200W
Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W
DJI Osmo Mobile 8 Gimbal with Mic Mini Transmitter, Infinity Black
As an affiliate, we earn on qualifying purchases.
From Cowork to Agent Race
The review begins with January, when NVIDIA presented its Vera Rubin platform at CES and Anthropic introduced Claude Cowork as a research preview. NVIDIA said Vera Rubin would deliver up to 10 times lower cost per token than its Blackwell platform. That is a vendor claim, and the source says partner systems were expected in the second half of 2026; it does not provide an independent performance assessment.
In February, Anthropic added Cowork plugins, including one for legal work. Bloomberg quoted a trader using the term “SaaSpocalypse,” while Reuters reported the software-index decline. The same month, Anthropic announced a $30 billion Series G at a $380 billion valuation. In March, the review records a dispute with the Pentagon over permitted uses of Anthropic’s models; a federal judge later blocked the government’s supply-chain-risk designation, according to the supplied account.
Later chapters cover model releases, open-weight competition and infrastructure spending. The source says the open-versus-closed model gap was six points in April, and that the output price of OpenAI’s Sol tier fell from $30 to $10 per million tokens across two generations. It also cites Big Tech capital-spending guidance of up to $725 billion for 2026 and an Anthropic valuation of $965 billion. These are reported figures in Meyer’s review, but the provided material does not identify all underlying documents or clarify the companies included in the spending total.
“A release packaging issue caused by human error, not a security breach.”
— Anthropic, describing the Claude Code source-map exposure, as quoted by VentureBeat
Evidence Behind the Year’s Claims
The review spans a fast-moving period and includes figures drawn from different sources. It says primary sources were checked where available, while acknowledging that some developments were reported by only one outlet. The supplied material does not include links or full documentation for every claim, so readers cannot assess all underlying evidence from this account alone.
Several claims require particular care. Anthropic’s account of alleged model distillation is the company’s own finding. The Vera Rubin cost comparison is NVIDIA’s claim. The purported Hugging Face server incident and the UK deception report are summarized without enough detail here to establish the precise test conditions or findings. The source also lists June releases of Claude Fable 5 and Claude Mythos 5 followed by a government letter and a shutdown, but does not include the letter itself or a detailed account of the controls.
The recap ends in October, not at year-end. Its forecasts, agent rollouts, spending estimates and security figures may change or receive further confirmation. The source’s prediction grades are Meyer’s own assessment; the supplied material does not include the full ten predictions or the evidence for each grade.
Year-End Tests and Follow-Up
The immediate next step is to compare the October account with the remaining months of 2026. Further reporting can establish whether the named agent products are widely deployed, whether their use changes software demand, and how companies handle the security risks raised by systems that can operate tools and access services.
Readers should also watch for primary documentation behind the review’s largest claims: official company filings and spending guidance, court records and government correspondence, technical reports on security tests, and independent assessments of model performance and pricing. Those sources would help distinguish demonstrated outcomes from forecasts, market reactions and vendor statements.
Meyer says the review includes links to selected coverage and evidence for the prediction grades. The supplied excerpt does not contain the full chapter links or all ten forecasts. Until those details and year-end results are available, the recap offers a structured account of developments through October, rather than a final verdict on AI’s impact in 2026.
Key Questions
What is the main development in this article?
Thorsten Meyer published a recap of AI developments from January to October 2026, arranged as ten monthly chapters and including a review of his January forecasts.
How did Meyer grade his January predictions?
Meyer says the ten predictions received two hits, five partial results and three misses. The supplied source excerpt does not list all ten predictions or the supporting evidence for each grade.
Did AI agents cause the software-stock decline?
The source links investor concern about AI disruption to a reported $1 trillion decline in the S&P 500 software and services index over seven losing sessions through February 5. That is a reported market move and attribution of investor concern, not proof that agents caused the decline or that software companies’ business models have been permanently damaged.
Which claims in the recap are independently confirmed?
The source cites Reuters, Bloomberg, the IEA, USGS and other material for selected details, but not every claim has the same sourcing. Company statements—including NVIDIA’s cost comparison and Anthropic’s account of alleged distillation—should be treated as attributed claims unless independently verified.
What remains to be seen after October?
The year-end record will show whether agent adoption, model pricing, security practices and infrastructure constraints develop as the recap suggests. Further documentation is also needed for some reported incidents and for the evidence behind the author’s prediction grades.
Source: ThorstenMeyerAI.com
