Article reader Listen + reading controls
Article reader
Preparing the reader…
Reading settings
Leadership for the AI Age Is Institutional Design¶
Calls for “ruthless” leadership in defense modernization usually express a valid frustration. Institutions can preserve legacy platforms, organizations, and processes long after their opportunity cost becomes strategically dangerous. Artificial intelligence and autonomous systems will not diffuse into military advantage through local enthusiasm alone; leaders must allocate resources and change how forces organize, train, decide, and operate.
But ruthlessness is not a strategy. Indiscriminate disruption can retire useful capability, concentrate risk in immature systems, and reward visible technology over the less glamorous infrastructure that makes it dependable.
Leadership for the AI age is better understood as institutional design under uncertainty.
The 2024 RUSI and Special Competitive Studies Project report Leveraging Human–Machine Teaming argued that the United States and United Kingdom should use human–machine collaboration and teaming to create decision advantage, operate in denied environments, and impose costs on adversaries. Contemporary commentary emphasized the need for determined civilian and military leadership and a willingness to reallocate resources.1
The enduring question is how leaders distinguish necessary transformation from fashionable churn.
Leaders Must Choose the Unit of Transformation¶
AI is not a discrete capability category like a new vehicle or sensor. It can change intelligence, logistics, planning, cyber operations, command and control, maintenance, training, and the design of autonomous systems.
An enterprise-wide “AI transformation” is too broad to manage. A model-centric initiative is too narrow to produce mission value. Leaders need bounded mission portfolios:
- a consequential operational or business outcome;
- a set of mission threads and users;
- the data, software, infrastructure, doctrine, and workforce dependencies;
- an accountable product or capability owner;
- and measures that compare the new system with the baseline.
This allows leadership to make real choices. A portfolio can be expanded, constrained, integrated, or stopped. A general commitment to AI can only be affirmed.
Divestment Requires Evidence and a Replacement Path¶
Legacy systems survive for reasons. They may be expensive and technically brittle while supporting essential workflows, certifications, training, interfaces, and organizational knowledge.
Retiring them requires more than declaring the future superior. Leaders should ask:
- Which mission outcome does the legacy capability support?
- Which dependencies are documented, and which remain tacit?
- What portion can be retired, encapsulated, or modernized incrementally?
- What evidence shows the replacement performs under operational conditions?
- What rollback or continuity path exists during transition?
- Which workforce and doctrine changes must occur in parallel?
The bold decision may be to terminate a platform. It may also be to fund the unglamorous reverse engineering and integration needed to remove it safely.
Divestment should release resources into a named replacement portfolio rather than a general innovation account. Otherwise, the institution experiences loss immediately and advantage only rhetorically.
Technology Budgets Conceal Enabling Work¶
Leaders are attracted to visible artifacts: autonomous platforms, foundation models, command applications, and predictive systems. The enabling work appears as overhead:
- data stewardship;
- identity and access;
- software delivery platforms;
- test and evaluation;
- observability;
- model and knowledge provenance;
- human-systems integration;
- cybersecurity;
- training and doctrine;
- and technical workforce development.
Underfunding these layers makes every program rebuild them locally or depend on a vendor’s proprietary stack. The result is a portfolio of prototypes with high marginal integration cost.
Senior leaders must fund shared foundations and require mission programs to use and improve them. Centralization should stop at the point where it erases mission ownership; decentralization should stop at the point where it recreates common infrastructure.
Decision Rights Are the Core Leadership Product¶
AI programs often stall because many organizations can object and few can decide. Data, cybersecurity, acquisition, legal, test, mission, and platform offices each hold legitimate responsibilities.
Leadership must define:
- who owns the mission outcome;
- who controls data and semantic definitions;
- who accepts which categories of risk;
- who authorizes limited and expanded use;
- who can suspend the system;
- who resolves cross-organizational conflicts;
- and which evidence each decision requires.
This is not a one-time governance chart. Decision rights should operate at product tempo. A board that meets quarterly cannot govern a system that changes weekly unless routine decisions are delegated within clear bounds.
The leader’s role is not to approve every model. It is to create an accountable system in which the right people can make timely decisions and material disagreements reach the appropriate level.
Learning Rate Is a Strategic Measure¶
Because AI and adversary behavior change, leaders cannot select a fixed destination and manage execution toward it. They need organizations that learn faster.
Learning rate depends on:
- time from mission hypothesis to representative test;
- quality of operational feedback;
- ability to observe technical and human performance;
- speed of safe software and model change;
- preservation of failures and negative results;
- and transition from successful experiment to funded ownership.
Exercises such as GIDE and iterative software practices are useful when they update architecture, doctrine, acquisition, and training—not when they merely generate demonstrations.
Leaders should reward teams for exposing invalid assumptions early. A culture that equates every pilot with success will learn slowly because failure becomes politically expensive to report.
Portfolio Governance Must Make Stopping Normal¶
AI attracts optimistic forecasts and senior attention. Programs can continue because termination appears to concede that the technology itself was overhyped.
A portfolio needs explicit stop conditions:
- insufficient mission improvement relative to the baseline;
- data or infrastructure cost disproportionate to value;
- unresolved safety, rights, security, or legal constraints;
- inability to transition from a special environment;
- unacceptable vendor dependency;
- or a stronger alternative.
Stopping one use case can free people and infrastructure for another. It is evidence of strategic focus, not opposition to AI.
The RUSI–SCSP report’s language of “ruthlessness” is most constructive when applied to portfolio discipline: protect neither legacy capability nor AI prototype from evidence.
Leaders Must Redesign Professional Identity¶
Human–machine teaming changes roles, expertise, and authority. Personnel may move from direct execution to supervision, exception handling, model evaluation, or policy design. New technical specialists enter command and acquisition relationships that were not designed around their expertise.
Leaders need to address:
- how operators maintain skill and situation awareness;
- how technical experts gain career status and decision influence;
- how commanders understand model limitations and data conditions;
- how responsibility remains human without making oversight ceremonial;
- and how education prepares leaders to challenge both automation bias and reflexive rejection.
Training cannot be postponed until after fielding. The organization and the technology must be designed together.
Accountability Must Accelerate With Capability¶
Faster sensing, analysis, and action can create advantage. It can also reduce the time available for deliberation, increase the scale of errors, and make causal chains harder to reconstruct.
Leadership must insist that speed is accompanied by:
- explicit human decision rights;
- provenance and auditability;
- bounded operating envelopes;
- tested degraded and failure modes;
- incident reporting and protected dissent;
- and the authority to disengage or roll back.
Responsible AI is not an ethical brake added to transformation. It is how leaders preserve command accountability as machines assume more complex roles.
The Strategic Inference¶
The AI age does require determined leadership. The difficult decisions are not limited to canceling legacy systems or funding new platforms.
Leaders must choose mission portfolios, fund invisible foundations, assign decision rights, build technical careers, create transition mechanisms, and stop both old and new systems when evidence fails. They must increase learning speed without allowing operational speed to outrun accountability.
This is less theatrical than technological revolution and more durable. The institution becomes capable of changing itself repeatedly as technology, missions, and adversaries evolve.
That—not any single model—is the transformation leaders must design.
This essay was substantially revised in July 2026 to replace the original short commentary with an evidence-based analysis of leadership and institutional design.
For related work on AI strategy, transformation systems, and technical leadership, visit my portfolio or connect on LinkedIn.
References¶
- Sidharth Kaushal et al., Leveraging Human–Machine Teaming, RUSI and SCSP, January 2024.
- Special Competitive Studies Project, “Report on Leveraging Human-Machine Teaming.”
- Sydney J. Freedberg Jr., “Transforming the military for the AI age requires ‘a certain ruthlessness,’ say US, UK experts,” Breaking Defense, January 18, 2024. This report was the historical prompt for the original post.
-
Royal United Services Institute, “Transforming the military for the AI age requires ‘a certain ruthlessness,’ say US, UK experts,” January 18, 2024. ↩