You’d never hire 5 people to write the same memo and pick the best one. But you should.
There’s a popular argument right now that AI makes us dumber. That we’re outsourcing our thinking. That the muscle atrophies.
I think the opposite is happening. And I think the people making that argument are confusing the tool with how it’s used.
Here’s what actually changed for me: AI didn’t reduce my cognitive load. It shifted where the load sits. The thinking got harder. It just moved from creation to judgment.
Over the past year I’ve been cataloging the things that are genuinely possible now that weren’t before. Not faster versions of old work. New cognitive moves that didn’t exist at human speed or human scale.
I’ve landed on six categories. This is my working hypothesis for what AI actually unlocks for knowledge workers.
1. Parallel Cognition at Scale
Last month I was building a client proposal for a capability development program. Instead of writing one version and iterating, I generated five different framings of the same engagement. Different structures, different lead-ins, different ways to present the pricing. My job was to look at all five and decide which one actually captured what I’d learned in discovery. Before, I developed one version until it either worked or I’d sunk too much time into it to walk away. Now I stress-test five framings in the time it used to take me to write one. The cognitive demand didn’t shrink. It moved from “can I produce something good?” to “can I recognize which of these is best and why?”
2. Time Compression
When I was drafting an MSA and statement of work for a new client recently, I went from blank page to finished contract in a single session. Draft, critique, rewrite, stress-test against edge cases, simplify. The iteration cycle that used to take a week of back-and-forth happened in hours. What I found is that faster iteration doesn’t mean less thinking. It means I have to evaluate faster. Every cycle demands a judgment call about what improved and what regressed.
3. Cognitive Load Elimination
I regularly feed AI a messy stream of half-formed thinking about a client engagement or a new framework and ask it to turn that into a coherent argument, then show me where I’m contradicting myself. It’s like a whiteboard that reasons back. Last week I was working through the structure of a capability model and it flagged that two of my core assumptions were in tension with each other. I argued back. Lost. I’d been holding that contradiction for weeks without seeing it. The thinking required to engage with that feedback, to actually sit with the contradiction and resolve it, was harder than the original idea generation.
4. Role Multiplication
Before sending a proposal, I now run it through multiple lenses: “Respond to this as the CFO who has to approve the budget. Now as the VP of Talent who championed it internally. Now as the skeptic on the leadership team who thinks L&D is a cost center.” I used to wait for those perspectives to surface in meetings. Now I stress-test before the meeting happens. One person, many perspectives, without the scheduling overhead. The demand on me is synthesis: holding conflicting viewpoints and deciding which objections to address preemptively.
5. Exploration Without Commitment
A few weeks ago I was designing a new go-to-market angle. Instead of committing to the idea and building it out, I deliberately pursued the version I thought would fail and traced exactly where it collapsed. Before, exploring a bad idea still consumed real time and real morale. Now the cost is close to zero, and the information I get from watching an approach break is genuinely useful. I learned more about what made the good version good by understanding precisely why the bad version didn’t work.
6. Meta-Work
This is the most underrated one. AI helps me design the workflow while I’m executing it. I’ve built a system that adapts to how I think, that knows to challenge me when I default to building frameworks instead of picking up the phone and selling. “Force me to decide instead of exploring endlessly.” The cognitive work here is self-awareness: being honest about your own patterns so the system can push against them.
The Thread That Connects All Six
Every one of these six moves demands more from me, not less. Parallel outputs require sharper judgment. Faster iteration requires faster evaluation. Structured thinking requires me to actually engage with the contradictions. Role multiplication only works if I can synthesize conflicting perspectives. Exploring dead ends only pays off if I know what to learn from the collapse.
The practices we built around knowledge work weren’t “best practices” in any absolute sense. They were adaptations to human limits. Brainstorming in a room together, single-threaded iteration, discrete review cycles spaced days apart. We called it process. It was really coping.
Those limits are gone. What remains is the hard part: judgment, framing, taste, problem selection, knowing what to discard.
The argument that AI makes us dumb assumes the valuable part of knowledge work was always the production. It wasn’t. It was the thinking that guided the production. AI just made that obvious by handling everything else.
I’m working through how to build these capabilities explicitly with the L&D leaders I work with. If you’re thinking about what development looks like when the old constraints disappear, I’d like to compare notes. DM me on LinkedIn.
