YoIf 2024 was the year of large language models (LLM), 2025 looks like the year of AI “agents.” These are quasi-intelligent systems that take advantage of LLMs to go beyond their usual tricks of generating plausible text or responding to prompts. The idea is that an agent can be assigned a high-level (possibly even vague) goal and broken down into a series of executable steps. Once you “understand” the goal, you can devise a plan to achieve it, just like a human would.
OpenAI CFO Sarah Friar recently explained it so to the Financial times: “I could be a researcher, a useful assistant for ordinary people, working mothers like me. In 2025, we will see the deployment of the first highly successful agents that help people in their daily lives.” Or it’s like having a digital assistant “who not only responds to your instructions but is able to learn, adapt and perhaps most importantly, take meaningful action to solve problems on your behalf”. In other words, Miss Moneypenny on steroids.
So why are these automated Moneypennys suddenly considered the next big thing? Could it have something to do with the fact that the tech industry has spent trillions of dollars creating colossal LLMs with, so far, no plausible return on that investment in sight? That’s not to say that LLMs are useless; For people whose work involves language, they can be really helpful. And computer programmers find them very useful. But for many industries, for the moment they still seem like a solution in search of a problem.
The arrival of AI agents may change that. Using LLMs as building blocks of virtual agents that can efficiently carry out many of the complex task sequences that constitute “work” in organizations around the world could prove irresistible. Or so the technology industry thinks. And so, of course, does McKinsey, the mega-consulting firm that provides the subliminal anthem that CEOs invariably sing. AI agent, McKinsey Bubbles“is moving from thinking to action” as AI-enabled “agents” that use basic models to execute complex, multi-step workflows are adopted in a digital world.
If that’s really what’s going to happen, then maybe we have to rethink our assumptions about how AI will change the world. At the moment, what we are most obsessed with is what technology will do to individuals or humanity (or both). But if McKinsey & Co are right, then the most profound long-term impact could come from the way AI agents change corporations (which, after all, are actually machines for managing complexity and turning information into decisions).
The political scientist Henry Farrell, an astute observer of these things, has suspected this possibility. LLM, he argues“They are engines for summarizing and making useful large amounts of information.” Since information is the fuel that large corporations run on, they will adopt any technology that provides a smarter, more contextual way to manage information – unlike mere data that currently process. So, Farrell says, corporations will “implement LLM in ways that seem boring and technical, except to those immediately involved for better or worse, but which are actually important. Great organizations shape our lives! As they change, so will our lives, in countless seemingly boring but meaningful ways.”
At one point in his essay, Farrell compares this “boring and technical” transformative impact of LLMs to the way the humble spreadsheet reshaped large organizations. This caused a gentle outburst by Dan Davies, economist and former stock market analyst whose book The irresponsibility machine It was one of the most pleasant surprises this year. He notes that spreadsheets “made a completely new style of work possible for the financial industry in two ways.” Firstly, it allowed the creation of much larger and more detailed financial models and therefore a different way of budgeting, compiling business plans, evaluating investment options, etc. And secondly, technology allowed us to work iteratively. “Instead of thinking about which assumptions made the most business sense and then sitting down to project them, Excel (Microsoft’s spreadsheet product) encouraged you to simply set the forecasts and then sit down and modify the assumptions up and down until you got an answer. I could live with. Or, at any rate, an answer your boss could live with.”
The moral of that story is clear. The spreadsheet was a revolutionary technology when it first appeared in 1978, as is ChatGPT in 2022. But now it is an integral and routine part of an organization’s life. The arrival of AI “agents” built from GPT-like models appears to follow a similar pattern. In turn, the organizations that have absorbed them will also evolve. And then the world might eventually rediscover that famous saying attributed to Marshall McLuhan’s colleague John Culkin: “We shape our tools and then the tools shape us.”
what i have been reading
Talking about economics
Transcription of a fascinating interview with the notable economist Ha-Joon Chang, on economics, pluralism and democracy.
AI or A-no
“The False Comforts of AI Skepticism” is a vigorous essay by Casey Newton on the two “camps” in AI arguments.
What Trump did next
“I have a cunning plan…” Charlie Stross’s blog entry is a sketch for a truly dystopian story on the aftermath of Trump’s inauguration.