More Capable, Less Able to Change Course?

A system may expand an institution’s operational capabilities while progressively eroding independently available routes for revising, replacing, or exiting the arrangement.

That possibility is easy to miss when technology is evaluated primarily through the tasks it improves. Better performance is visible: more cases processed, more inconsistencies detected, faster coordination. The loss of alternatives may remain unnoticed until someone tries to challenge a decision, change a workflow, replace a provider, or stop using the system.

Consider a hypothetical public procurement team adopting an AI-supported platform. The platform helps staff compare bids, identify conflicting claims, and coordinate reviews. Suppose a supplier tries to redirect an evaluation through conflicting submissions and pressure to relax the agreed criteria. With the platform’s support, the team reconstructs the evidence, responds to the challenge, and sustains those criteria where it previously could not.

Over time, the platform becomes the working environment. Records accumulate in its formats. Review procedures assume its availability. Staff increasingly specialize in supervising its outputs. Independent tools and procedures receive less attention.

None of this necessarily creates a problem. Specialization and shared infrastructure can support valuable capabilities. The question is what happens to the alternatives.

Suppose the team later needs to change its evaluation approach or replace the platform. It discovers that the transition requires reconstructing records, recovering neglected skills, and keeping ongoing procurements moving. Its formal authority to make the change remains intact. The relevant technical controls still work. A timely transition may nevertheless be beyond its available resources.

Two trajectories have diverged: the team’s ability to perform within the existing arrangement has expanded, while its practical ability to revise that arrangement has contracted.

This distinction helps clarify the relationship between agency and autonomy.

We use agency in a stronger sense when discussing the ability to pursue or defend goals under active opposition. Higher productivity alone does not establish such an increase. In the hypothetical example, the relevant improvement is the team’s ability to sustain its evaluation against attempts to redirect it. An empirical finding of stronger agency would require evidence of that improvement and of the system’s contribution to it.

Autonomy concerns the ability to form, accept, reject, and revise goals, criteria, and the conditions or terms under which one acts. A larger selection of tools for carrying out an assigned task does not necessarily expand that ability.

The relationship is therefore specific. With support from a system B, an actor A may become better able to withstand opposition from C, while losing practical ways to renegotiate its relationship with B.

Independent alternatives are one condition supporting autonomy. They do not exhaust its meaning. Their importance depends on whether they enable a person or institution to reconsider goals, change the terms of participation, or leave an arrangement while preserving necessary functions.

Independence is also relative. An alternative may rely on another provider, a professional association, a public institution, or a coalition. It need not involve doing everything alone. What matters is whether the route depends on the particular cooperation, permission, or resource whose availability is in question.

These distinctions become especially important when assessing control.

Four questions concern whether an intervention can be exercised, before it is attempted:

1. Does the actor have authority to intervene?
2. Does the intervention mechanism work?
3. Are the information, access, skills, resources, and coordination needed to execute it available?
4. Can the intervention preserve necessary functions within the relevant limits?

A separate question concerns evidence of actual exercise: Was the intervention carried out, and what result followed? The first four questions assess conditions of exercisability in advance; this last question concerns what happened in practice. A feasible intervention need not have been exercised, and its eventual result must be established separately.

A functioning shutdown mechanism answers only part of this assessment. Effective control requires an available route through the relevant conditions.

We call this practical exercise feasibility: the availability of a plan that can achieve a specified control objective within the relevant constraints.

The plan may involve correcting one decision, limiting a system’s permissions, suspending part of a process, transferring a function, or replacing a provider. Complete shutdown is only one possible intervention.

For any given objective, we can consider a set of candidate plans and identify the subset that is practically feasible. Dependence can remove plans from that subset without changing the formal rules or the technical mechanism of control.

The limiting factor need not be money. It may be a missing skill, incompatible data, an exhausted reserve, an unavailable third-party permission, or insufficient time to maintain continuity during a transition.

This problem overlaps with vendor lock-in and exit planning, but the mechanism considered here is narrower. Feasible control routes may disappear without opportunism, threats, resistance, or any change in formal authority. The constraint can arise from the surrounding organization of activity. A route’s existence must also be established rather than inferred from a nominal exit plan.

This makes practical control an architectural question. The assessment must follow the dependencies around an intervention, including those of its supposed alternatives. A backup that relies on the same unavailable service may help with one kind of failure and offer no protection in the scenario that matters.

The evidence also needs careful limits. Failing to identify a feasible route does not prove that none exists. An assessment should distinguish routes supported as feasible, routes excluded by established constraints, and routes whose feasibility remains unknown.

A defensible finding might be:

Within the defined search space, time horizon, and scenario, no feasible intervention route was identified.

A stronger observation becomes possible when the same route is assessed over time: it was previously available, and a documented change in dependencies has made it unavailable. Losing the records needed to assess that route is a different result.

The mere number of alternatives can also mislead. Several new routes requiring the same provider’s discretionary approval may offer less independence than one route that remains usable without it. Comparison must preserve the functions and conditions that make an alternative matter.

The divergence described here is a research hypothesis, not an inevitable technology lifecycle. Systems can expand operational capability while preserving or increasing independent alternatives. An institution may retain compatible records, maintain skills outside the platform, and sustain workable transition arrangements. Whether these measures are sufficient requires evidence in the relevant scenario.

The empirical task is to establish the changes separately: which capabilities improved, which alternatives were lost or gained, and what connects those developments. Continued use alone establishes neither trust nor the absence of alternatives. An unused intervention may still be feasible. A formal exit right may be practically difficult to exercise.

This returns us to When Political Concepts Become Machine-Readable. Formal authority and mechanism functionality may be comparatively straightforward to encode as criteria, though neither is self-interpreting. The availability of skills and resources, preservation of necessary functions, counterfactual transition costs, and feasibility under constraints require external evidence and organizational knowledge. A machine-readable assessment must preserve the distinction between an established condition and an unexamined assumption.

Who should assess the controllability of a system when the evidence needed for that assessment is partly produced, structured, or mediated by the system itself? The harder control becomes to operationalize, the greater the risk of allowing the controlled system to define what counts as evidence of its own controllability. The assessment therefore needs grounds for challenging both the evidence and the criteria through routes that do not depend entirely on that system.

For Indispensable Systems, this changes the unit of evaluation. A system’s contribution includes both the activities it enables and its effects on the institution’s ability to reconsider the arrangement.

The question is:

Which capabilities does the system expand, which independent routes for revision and exit does it preserve or narrow, and through what changes in dependence do those trajectories diverge?

An organization’s ability to keep operating and its ability to change course deserve to be examined together.