The productivity trap

Companies treating AI as a productivity tool get real gains but lose them if they don't change processes and incentives. More PowerPoint and code without strategic purpose wastes the opportunity.

“If you don't change any part of the process, AI is mostly a single player entity and you end up with more of everything.”

AI can do two to eighteen weeks of work

AI regularly completes work that would take a human two to eighteen weeks in reasonable time and money with high success rates. This changes what organizations must do.

Jobs are blurring

As AI capabilities expand, job boundaries blur. Coders do design work, designers code, product managers do everything. Leaders must rethink organizational boundaries and what the business actually produces.

“Jobs start to blur. Coders do design work, designers do coding, product managers do everything.”

Leadership, lab and crowd model

Mollick recommends three things: leadership making decisions about AI use and incentives, crowd access to current tools, and a lab doing twenty-four-seven work on organizational restructuring around AI.

Leaders must eat risk

Leaders who succeed with AI are willing to experiment actively without letting organizational vetoes block progress. They use the latest systems themselves rather than waiting for reports or training.

“Unless we do active experimentation, we're not going to know anything.”

From centaur to conjurer

AI work has evolved from half-human, half-AI collaboration to assigning AI tasks like team members. A conjurer writes an RFP or SOP and the AI executes the work autonomously with high success rates.

“You are actually assigning the AI to tasks almost like your team you're assigning and it does the work.”

Everyone becomes a manager

Everyone now manages AI. Management skills look like AI skills. Studies show the best coders at first pass are managers who specify what they want and let AI execute, not coders themselves.

Twilight factories, not dark factories

StrongDM ships product with no human involvement, but Mollick warns this is dangerous. Instead, build twilight factories where AI proactively brings humans in for expertise, sign-off and interesting work.

Leaders stay empowered

Leaders managing swarms of a hundred agents get PowerPoint reports and demos rather than approving every action. They can query company information directly and diagnose problems no one brought to them.

AI makes strategic decisions

Epoch reports getting two to eighteen weeks of engineers' work from a single prompt to Claude for a couple hundred dollars.That involves strategic decisions and judgment calls.The question is where you want AI deciding.

Four capabilities matter most

Deep knowledge of your field, wide knowledge across fields, taste to pick among options, and agency to act under uncertainty. These become increasingly valuable as AI handles routine work.

Experience still wins

Studies with BCG consultants and Anthropic show less experience results in worse AI outcomes. Senior leaders with expertise and judgment have the best tools available to make decisions about AI use.

Use frontier models for strategy

Leaders should get second opinions from frontier models like Claude Fable five point one or GPT-six on strategy, legal and medical advice. The only way to know what's good or bad is by using it.

Large companies can compete

Large companies have one hundred thousand people, the smartest people in the world, customers and resources. Startups lack experience and the same barriers. Everyone uses the same AI models.

Startups have never had it better

Startups are top zero point one percent at one thing and terrible at everything else. AI makes them good at everything they weren't good at, solving problems they couldn't solve before.

Cybersecurity will get intense

OpenAI pulled twenty percent of resources to hardening systems. Large financial institutions are pulling fifty percent plus of staff. Agents will relentlessly attack systems for benign reasons, creating chaos.

Stop fixating on today's capabilities

The biggest mistake is anchoring on what AI can do now when capabilities change every three months.Build for what AI will be capable of, not just what it can do today.Stay in the exponential.

Build capacity in-house

Every vendor and consultant is making this up as they go. Build the capacity to make your own solutions in-house. Don't trust that other people have this nailed down or have successful cases.

“Every vendor you talk to is also making this up as we go along.”

Ambidexterity is essential

Exploit what's happening today while exploring the future. Ask what you've stopped doing, what impossible thing you're building, and whether you have leadership, lab and crowd in place.

Culture and incentives decide everything

If your organization has good structure, culture and reward systems, people discover use cases and tell you. If not, they hide everything because they're scared. Incentives determine what people do with AI.

“People do what they're incentivized to do.”