AI & Strategy

Two Shifts Right Now

• The exponential curve continues. AI keeps getting better — no sign of slowing.
• Agents are real. A single prompt can now yield 2–18 weeks of expert work at high success rates (per Epoch benchmarks), for a few hundred dollars.

The Productivity Trap

Treating AI as just a productivity tool means gains get eaten by the organization — more PowerPoints, more code, same KPIs. Default to automation and you end up with "the same Claude-based company everyone else does."

The real question is not "Where can we use AI alongside humans?" but "How does AI change our business?"

Jobs are blurring. Coders do design. Designers code. PMs do everything. Leaders must rethink organizational boundaries, outputs, and goals — urgently.

The Winning Model: Leadership + Lab + Crowd

All three are required. Leadership without a Lab is talk. A Lab without the Crowd misses the real use cases. A Crowd without Leadership produces chaos — or silence.

Leaders who win "eat risk." Experiment actively. Use the tools yourself. Letting everyone veto because "it might be risky" is itself risky.

Conjurer & Twilight Factories

From Centaur/Cyborg → Conjurer. You no longer prompt-and-review. You conjure a team of agents; prompting now looks like writing an RFP or SOP. Everyone is a manager. Anthropic found the best first-pass coders are managers — they specify well; AI executes.

• Dark factory = 100% AI, no humans (e.g., StrongDM shipping code with zero human review). Risky, undifferentiated.
• Twilight factory = AI proactively pulls humans in for expertise, variance, judgment, or sign-off. Human + AI + process + culture = competitive advantage.

Four Capabilities to Develop

1. Deep knowledge — spot a bad answer in seconds.
2. Wide knowledge — cross boundaries, prompt better.
3. Taste — choose among many AI-generated options.
4. Agency — act under uncertainty.

Senior leaders have an edge. AI isn't a "digital native" game. Studies (including Anthropic's and early BCG research) show less experience = worse AI outcomes. Expertise, judgment, and comfort with uncertainty are advantages.

Four Diagnostic Questions

1. What have you stopped doing because AI made it trivial?
2. What are you building that's impossible — betting the curve keeps growing?
3. Are you getting a second opinion from a frontier model on major decisions?
4. Do you have Leadership, Lab, and Crowd in place?

Two Biggest Strategic Mistakes

• Fixating on today's AI capabilities. The "hot thing" changes every three months.
• Trusting someone else has figured it out. Vendors, consultants, even AI labs are making it up. Build internal capacity.

Risks on the Horizon

• Cybersecurity: expect a near-term "apocalyptic" environment — agents relentlessly probing systems, often for benign reasons. OpenAI reallocated ~20% of resources to hardening; some banks 50%+.
• Jobs: nobody knows. AI labs explicitly aim to replace human labor. "We shouldn't let that be the future we get."
• Augmentation vs. automation: if humans only verify AI output (and get fired when it's wrong), that's a terrible way to work.

"People do what they're incentivized to do. If your incentives are 'minimize the impact of AI' or 'use AI to do more of what my KPI says,' it's not transformation."

— Ethan Mollick