How to Govern AI While Still Capturing the Payoff
Integrate AI to multiply competence without birthing a coordination bottleneck
The Coordination Problem
The integration of AI increases competence but can also spike coordination efforts.
Consider the following examples:
Increased work - AI, for instance, helps job applicants prepare for interviews—but it also helps recruiters raise the intensity of desired qualifications and evaluations. All things considered, the job application process has created a bottleneck for the job market. The same phenomenon is manifesting with cybersecurity.
Increased bureaucracy - AI integration thrives when companies consolidate on platform and tooling. Yet, the uniform integration of AI is most effective when the task is distributed directly to the overseer of an organizational slice, allowing them to restructure their teams and systems around the capabilities of AI—rather than weaving AI into assumed structures shaped by pre-AI capabilities.
Increased risk of over-engineering - The AI race between America and China has a ripple effect on other nations. The UK is taking a protectionist approach. The EU has mobilized to enact an AI Act. Global companies are forming an alliance on cybersecurity. In 2025, American companies rushed to adopt AI early to gain an edge. Exploring software engineers have obsessed over harness engineering techniques around prompting, skills, agent swarms/organizations, and the like. Yet, frontier models continue to rapidly evolve and absorb the work that we expected from human organizations—and the latest research only predicts that models will accelerate exponentially.
Any attempt to marshall human organizations to optimize around AI capabilities quickly becomes outdated and ineffective as models evolve.
The need to move fast also doesn’t create the conditions for imagining how human organizations may be assuming structures optimized for pre-AI capabilities.
Organizations need to swiftly mobilize as AI capabilities evolve, and that requires discerning how to effectively tear down (and not just set up).
Given this, AI governance is most effective when lightweight and minimalistic, creating seamless reporting and restructuring mechanisms.
Otherwise, the potential for payoff will be obscured, and the will to integrate evaporates.
The Coordination Solution
Here are rules of thumb to maintain lightweight and minimalistic AI governance to thwart increased work, bureaucracy, and risk of over-engineering:
Consolidate on minimalistic platform and tooling
Example: Claude Code licenses for all is good strategy even if there are greater opportunities for cost-optimization with the nitty-gritty of harness engineering.
Pin agents to people rather than creating agent organizations mediated by a central team
It is much easier for accurate reports to be gathered from personal testimony of agent use rather than complex calculations of the costs and value of central agents-as-a-service teams.
Make leaders of organizational slices directly responsible for reporting AI integration, eliminating unnecessary middlemen
Strategy needs to come from the top, but it doesn’t require a bloated middle
Don’t make every domain utilize AI
HR’s heavy use of AI has only created stagnation and breeded cynicism, for example
Only domains that heavily rely on competence (and not coordination)—such as software engineers—should integrate AI
Blindly throwing AI at teams is the new man-month fallacy
Invest in researchers/strategists around AI integration
Have them mine the latest AI news and research as a full-time activity
Give them a direct seat at the executive roundtable
Re-deploy resources; Don’t displace them
Workers won’t easily adapt if they sense a risk of displacement—it’s foolish to try to fool people
Coordination needs to set up and tear down swiftly as AI capabilities evolve.
A culture where workers can be re-deployed to new strategies will invigorate your best talent.
It also allows companies to coordinate around the current and anticipated states of AI capabilities.
Pre-AI, companies could afford to blindly layoff without significant pushback.
Post-AI, trust is paramount, and social costs are costly. The job application market is also incredibly ineffective in the present, and scaling up cannot be effective with a revolving door.
In short, organizations should govern AI because blindly throwing AI won’t reach a payoff.
But governance that squares itself with a post-AI reality (instead of assuming pre-AI strategies and structures) can reach the prized payoff.


Discussion about this post