With a new year upon us, we’ve been thinking about how far technology has come and where it needs to go. Here are three developments we’re thinking about.
Large language models have improved significantly for real-world work tasks.
One way to see how much AI has improved this year is GDP value (opens in a new tab)a benchmark developed by OpenAI that measures how well today’s LLMs can perform a range of real-world work tasks.
The benchmark asks human workers and LLMs to complete tasks designed by experienced professionals. He then asks another professional to choose which results were best, without knowing who did what.
OpenAI’s GPT-4o LLM, released in May 2024, beat or matched human professionals 12% of the time. Anthropic’s Claude Opus 4.1, released in August 2025, achieves near parity with humans, matching or outperforming professionals nearly 48% of the time. OpenAI claims that its new model, GPT-5.2 Pro, beats or binds professionals (opens in a new tab)about 74% of the time.
That doesn’t mean OpenAI’s latest model can do three-quarters of all tasks, and it certainly doesn’t mean it will replace 74% of jobs. The OpenAI framework intentionally skews toward digital work, and performance on generic, clearly defined knowledge-based work tasks is different from performing them in a company or within a team.
What this suggests, however, is that GenAI is poised to greatly disrupt knowledge work – and is making rapid progress – even though we are only in the early stages of that disruption.
The impact of GenAI on employment is still largely invisible unless you are early in your career.
Last year, Stanford’s Nicholas Bloom predicted (opens in a new tab) that 2025 would be the “year of the duck” for remote work: the number of people able to work from home would appear stable on the surface, while local battles over return-to-office policies would play out underneath.
This turned out to be an apt metaphor for AI and economics as well.
Several studies reveal that GenAI does not yet have an overall impact on employment. A analysis (opens in a new tab) Yale’s Budget Lab found that the share of jobs in the economy most exposed to AI has remained stable. A working document (opens in a new tab) Researchers at the University of Chicago found that people with high exposure to LLMs have experienced stable employment since ChatGPT’s release, while their salaries have increased. And asked this month about the role of AI in the weak labor market, Federal Reserve Chairman Jerome Powell. said, (opens in a new tab) “It’s probably part of history. It’s not yet… a big part of history.”
Likewise, many executives say AI has yet to make a significant impact in their business. In November, almost 90% of UK and US executives surveyed said AI had had no impact on employment at their companies in the previous three years, according to research. data (opens in a new tab) collected by researchers at the Bank of England, the Federal Reserve Bank of Atlanta and other institutions. Roughly the same proportion said AI had no impact on their company’s productivity.
Things have been less calm beneath the surface.
This year has seen a wave of memos (opens in a new tab) Tech CEOs say their companies are AI-driven, with some saying teams should prove AI can’t do the job before asking for more people or resources.
A few months later, we found that companies had given up on hiring entry-level workers. A working document (opens in a new tab) Stanford researchers found that employment for entry-level positions in jobs most vulnerable to AI automation — such as software developers and customer service representatives — has declined significantly since ChatGPT’s release. Meanwhile, employment for higher positions in these occupations has remained stable or increased. Another working document (opens in a new tab) Harvard researchers found that employment in junior positions with high exposure to AI fell following companies’ adoption of the technology.
We will be closely watching whether AI will have a broader effect on the job market and economy next year. The same sample of British and American leaders who said (opens in a new tab) AI is not yet impacting employment or productivity in their company and they expect this to change over the next three years; more than 40% of U.S. executives surveyed expect AI to lead to a reduction in workforce.
GenAI reduces friction – and that’s not always a good thing.
One of the hottest AI buzzwords of the year was “workslop,” a term invented (opens in a new tab) by researchers at BetterUp and Stanford to refer to AI-generated content that “presents itself as good work but lacks the substance to meaningfully advance a given task.” This shifts the cognitive burden from the creator to the recipient, who must now edit the generic, low-quality work. Around 40% of professionals surveyed said they had received one form or another in the previous month.
But labor issues are only part of a larger problem that has come to light this year: AI reduces friction, which can serve an important purpose.
Before AI, workers had to struggle to write a memo or strategy document, forcing them to think about the problem at hand. If they didn’t know something, they researched it, talked to a colleague, or looked deeper into the problem. With AI, it’s easy to skip these steps and produce a document that’s passable, or at least appears to be passable at first glance. The problem becomes worse if AI-generated documents are used to inform future decisions.
A similar dynamic is playing out in the labor market. Technologies like the “Easy Apply” button on LinkedIn already made applying for jobs easier, and AI was exacerbating the problem (opens in a new tab)allowing candidates to easily tailor their cover letters and resumes to individual jobs and automatically apply to hundreds of positions, regardless of their level of interest.
GenAI removes a significant source of friction: the work required to personalize and write applications. New search (opens in a new tab) suggests that in doing so, AI makes it harder for recruiters to identify good candidates from the pack.
Companies will need to rethink how they evaluate candidates in this new world. Financial Times columnist Sarah O’Connor has developed a good list (opens in a new tab) some of the solutions job platforms and employers are considering, including scheduling in-person or phone interviews earlier in the process and using AI-proof online assessments.