OpenAI's Dots and the Agent Race: What AI Is Actually Doing to Jobs Data

OpenAI used its annual DevDay developer conference in San Francisco on Tuesday, 29 September, to launch Dots, described as "always-on" agents for tasks such as booking travel, paying bills and replying to emails, according to CNBC. The launch came three weeks after Meta released its Muse agent earlier in September, a period in which Meta stock rose 29% on the month, its best since 2013, CNBC reported.

That is the news peg. The harder question is what any of this is doing to jobs.

On the same day, The Register published a column by Corey Quinn arguing the battle between OpenAI and Anthropic is now about distribution, not model quality. Quinn wrote that OpenAI is making its Codex harness open-source and adding Baseten to a marketplace of 32 launch partners, which he reads as a signal that OpenAI wants its models to "win on their merits." Anthropic, he wrote, takes the opposite line: try using a Claude Pro account with OpenCode and "you will shortly find that you no longer have a working Claude Pro account." That is a column, not a company statement, and the terms Quinn describes are only visible inside enterprise OpenAI accounts.

Jobs data splits in two directions

On the employment side, the freshest hard numbers in the dossier are not about AI at all. E2, using US Department of Energy data, found the US lost 36,949 clean energy jobs in 2025, the first annual decline since the pandemic, Electrek reported on 28 September. Clean energy employment fell to 3.52 million. Losses hit 35 states; California shed nearly 21,000 jobs, while Florida gained about 3,800.

That is a policy story. But it sits inside a wider labour picture that several outlets have been trying to read. The broad US energy industry lost an estimated 86,000 jobs in 2025, with clean energy accounting for roughly 43% of the fall, according to the DOE data E2 cited. E2's full Clean Jobs America 2026 report, with state and county detail, is due in October, so the current figures are initial. The group also recorded 142 clean energy manufacturing, generation and storage projects cancelled or downsized in 2025, and was careful to note that the employment figures do not establish how many jobs were lost to any single policy change.

The AI jobs debate is noisier, and much of it is projection rather than measurement. The dossier's recent headlines include a McKinsey estimate that 11 million US workers may need new jobs by 2035, a TD Economics report calling an AI "job apocalypse" unlikely, and a Revelio Labs report (via PR Newswire) saying the US added 56.9k jobs in September while the pace of new AI adoption fell 48% from its spring peak. Those are headline summaries, not full texts, and they point in different directions: adoption cooling, job growth continuing, and a long-run reshuffle still forecast.

"I think what we are watching is an inevitable evolution of the technology," said John Waldmann, CEO of HR tech startup Homebase, in CNBC's report. "It's probably no surprise that it's all happening around the same time."

CNBC's piece also notes the competitive backdrop: Google markets Gemini Spark as a "24/7 personal AI agent," Apple is pushing a redesigned Siri, and startup Instinct raised $1 billion at a $10 billion valuation on Monday. Town CEO Jean-Denis Grèze told CNBC that OpenAI's Dots "continues to validate the category and the need for work assistants." Grèze's comment is a statement to CNBC, not an independent finding.

Where the numbers are soft

If you want a single reason to be sceptical of any clean AI-jobs number right now, look at the timing. The most concrete AI-adjacent labour data in the dossier are monthly and noisy: US jobless claims fell to their lowest level since July, and ADP reported private sector job gains in September, per The Economic Times and WIOD. Investopedia's preview of Friday's labor market report notes that the picture is still forming. None of that isolates AI as a cause.

The same caution applies to the clean energy numbers. E2's finding is a genuine first: four straight years of growth ended in 2025. But it is a 2025 total, published in late September 2026, and E2 itself frames the result as the direct impact of federal support being rolled back, not as an AI effect. Analysts who want to tie AI to job losses need a different dataset.

There is also a structural point buried in the CNBC story. The personal-agent market now has OpenAI, Meta, Google, Apple and funded startups all shipping similar products at the same time, with, as CNBC put it, "little ability to lock in users." That is a product-market problem, but it is also a labour-market signal: when five large companies converge on the same assistant category within weeks, the work of junior coordinators, schedulers and inbox managers is squarely in the target zone.

What the dossier does not contain is a credible measurement of that displacement. The McKinsey 11 million figure is a forecast to 2035. TD Economics argues against an apocalypse. Revelio Labs reports both job additions and a fall in AI adoption pace. Those can all be true at once, and none of them is a payroll count.

For now, the only thing clearly dated and clearly sourced this week is the product race: Dots on 29 September, Muse earlier in September, Instinct's raise on Monday. The jobs verdict is still pending, and the next real datapoint is Friday's US labor market report, with E2's fuller clean energy breakdown due in October. Until those land, treat every AI-jobs headline as a projection wearing a number.