I don't really trust social science, what with their almost complete inability to replicate any results and endless stories about cooked data, but I always get a kick of out it producing a result I like. The science is this recent study, really a survey analysis, of knowledge workers self-reported experiences using AI for work. The title gives us a nice hint, with its 'reductions in cognitive effort', but the abstract has a quote that had me grinning like a madman:
Quantitatively ... higher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking. Qualitatively, GenAI shifts the nature of critical thinking toward information verification, response integration, and task stewardship.
That's right: the more people trust AI, the less they are likely to think critically about their work. And even when they do, their critical thinking is mostly within the confines of what a model has given them: they verify it, incorporate it into their work, and continue on down their task list.
I can't judge these knowledge workers too harshly. Earlier this week, my boss put together a Retrieval-Augmented Generation tool including all of HubSpot's CRM API documentation: that basically means that we get all of the logic of an LLM and a whole lot more certainty that its answers will be right, because it runs everything through what we told it is the canonical data source for these types of questions. To say that I rely on this is an understatement: it's easily ten times more efficient at one of my most common, most technical tasks, showing people the way to use our CRM APIs for their objectives. I still do some doublechecking and still incorporate the outputs into broader documents or decks, but that seems exactly like the "verification" and "task stewardship" that our social scientists found the AI-trusters doing.
What do we do when a frequent element of our daily work becomes that much faster? What do we do when something that used to take a significant amount of brain power becomes so much easier? As individuals, we immediately bank the win and move on to the next piece of work; as organizations, after a little while, we'll ask people to do a bit more. The more interesting layer is in the longer term how this kind of radical efficiency will adjust the kinds of roles that exist. Roles might become that much deeper, where we expect people to do more work along the lines of what they already do. Deeper roles could be things like tech support reps being both frontline and backline simultaneously. I think that's less likely than roles becoming wider: expecting people to wear simultaneously multiple people's hats. Maybe the frontline support person also acts as a frontline salesperson.
I read with interest a piece from the folks at Clay (an AI software concern) about how they condensed business development, sales rep, and sales engineers/architects into a role type they are calling go-to-market engineer. You could think of it as the sales world's equivalent of engineering's "full stack developer." They expect the approachability and output of AI tooling to allow people to run an entire sales funnel solo: prospecting, working deals, and providing technical proofs of concept. Like any piece of corporate content, it's a little bit of talking their own book: of course their GTM engineers use the heck out of a nifty software sold by Clay. I think there is something real there, too.
This week's reading: doing your job with AI.
How we're revamping traditional sales roles like SDRs, AEs, and sales engineers into a single, high-leverage function.
The career advantage nobody's talking about (yet)
The Index’s initial report provides first-of-its-kind data and analysis based on millions of anonymized conversations on Claude.ai, revealing the clearest picture yet of how AI is being incorporated into real-world tasks across the modern economy.