When we first started talking to the Big Four about AI, the conversations moved surprisingly fast. There was no long introduction, no need to explain why AI mattered. Almost immediately, partners and senior consultants began talking about where their time was actually going. Not into decision-making or insight, but into preparation. Pulling data from different systems. Aligning numbers. Making sure analyses matched internal standards before anyone could even start thinking about what the results meant.
What struck us was that none of this work was low value. It was necessary. But it was also exhausting. Highly experienced people were spending a large part of their day just trying to get to a point where real thinking could begin.
As the conversations continued, they mapped out more than dozens of processes where AI could play a role. Reporting, compliance, financial controls, tax analysis, internal reviews. On the surface, it looked like a classic automation problem. It wasn't.
Most of these processes already had tools, templates, and partial automation in place. What slowed everything down lived between the steps. Switching contexts. Reconciling outputs. Applying judgment consistently across fragmented information. The real cost wasn't time alone, it was cognitive load.
That's when the framing changed.
Instead of asking how AI could do the work end to end, we started asking how it could support the people doing the work. What if AI could handle the parts that drain energy, without taking away control? What if it could prepare structured drafts, surface inconsistencies, and make its reasoning visible, so experts could focus on judgment rather than setup?
Once this approach went live, the impact became tangible.
Across teams, cycle times dropped by 80 to 90 percent for recurring analytical work. Manual errors, especially in reconciliation-heavy processes, fell sharply. Just as importantly, outputs became more consistent across teams that used to approach the same work in slightly different ways. On average, consultants regained around 40 to 50 hours per month that could now be spent on higher-value thinking.
But the most interesting change wasn't in the numbers.
It was in how people talked about their work. Reviews became calmer. Discussions became more focused. Senior reviewers spent less time checking whether everything was correct and more time discussing what the results actually meant.
Over time, the system stopped feeling like a tool that people had to use. It started to feel like a shared way of working, quietly absorbing how decisions were made and reflecting that back in future analyses.
This experience also reinforced something we've believed for a long time. Progress toward more general intelligence won't come from systems that try to replace experts in complex domains. It will come from systems that reduce friction around expert thinking and amplify it.
In these engagements, AI created value not by being autonomous, but by being collaborative. It didn't make decisions for people. It helped them make better ones, faster.
If there's an early signal of where AGI-like systems will matter first, it's here. Not in full automation, but in supporting human judgment at scale.
That shift, from automation to leverage, is what made the difference.
