Introduction

I’ve previously discussed the likely first-order and second-order effects of AI on the Australian economy.

Here, I consider how AI may affect the policy environment itself. I outline a set of plausible propositions about how policy-making may change, along with the kinds of observations that would tend to support or weaken each.

TL;DR

If we think of the policy pipeline as: 1. Problem identification 2. Option generation 3. Analysis 4. Consultation / review 5. Decision 6. Implementation

Then AI is likely to strongly accelerate:

(2) option generation
(3) analysis

partially accelerate:

(4) consultation processing

and only weakly affect:

(5) decision
(6) implementation

The result is a shift in constraint — as the supply of policy options increases, the capacity to evaluate and select among them becomes more salient.

Possible implications

increased influence of actors who shape what is considered (e.g. ministerial offices, senior officials, central agencies);
a shift in lobbying toward access, credibility, and timing rather than analytical production;
faster iteration and tighter feedback loops in policy development, without a proportional increase in final decisions;
growth in policy complexity, particularly through more conditional and fine-grained rules;
an expansion in the range of proposals that are treated as policy-legible; and
increased difficulty in converging on decisions in some domains.

The Framework: AI favours agenda-setting and attention management

Policy proposals that receive serious consideration have traditionally emerged from a relatively structured set of processes: political and bureaucratic development, supplemented by contributions from academic, industry, and advocacy groups. These proposals typically took well-understood forms—white papers, committee reports, and similar artefacts—which served both as tools for developing ideas and as informal gatekeeping mechanisms. Participation required time, expertise, and institutional access.

Social media has already altered this landscape by enabling new forms of political pressure and agenda formation outside these traditional channels. AI extends this shift by reducing the cost of participating in the more formal, policy-shaped aspects of the process.

In particular, AI reduces the cost of:

drafting;
modelling; and
generating policy options.

As a result, a larger number of proposals can be produced in forms that meet the expectations of the policy process. The number of proposals that are sufficiently well-formed to merit consideration is therefore likely to increase.

At the same time, the capacity of political and administrative processes to evaluate and enact policy does not expand proportionally. Time, attention, and accountability constraints remain.

The result is a shift in constraint. As the supply of policy options increases, the ability to evaluate and select among them becomes more salient.

The effect of AI is best understood not as making policy easier to produce, but as shifting the bottleneck — from production to attention and coordination.

Effects

1. Agenda-setting power increases

Not all of these proposals can be considered. The process by which some are selected becomes more consequential.

In an environment where many proposals are sufficiently well-formed, not all can be evaluated. The policy process therefore depends more heavily on mechanisms that determine which proposals enter active consideration.

This increases the importance of actors who shape that selection. In practice, this includes ministerial offices, senior officials, central agencies, and other trusted intermediaries who structure the flow of information within government.

The effect is not that fewer ideas exist, but that a relatively small subset receives sustained attention. The selection of that subset becomes more consequential.

Implications

influence becomes more closely tied to access, trust, and proximity to decision-making processes;
framing and timing increase in importance, as they affect whether a proposal enters the decision set at all; and
agenda-setting influence becomes more concentrated, even as the range of possible proposals expands.

Testable Results

increased reliance on internal filtering and prioritisation before proposals reach senior decision-makers;
greater importance placed on briefings, summaries, and curated inputs;
a widening gap between the range of proposals discussed publicly and those treated as “serious” within institutions; and
increased value placed on relationships and credibility as pathways into the decision process.

2. Compliance cost falls; complexity rises

As the bureaucracy and industry adapt, it is plausible that greater complexity in policy can be generated and implemented. This can take both positive forms, in the sense of increased nuance, and negative (distortionary) forms, where complexity does not improve outcomes or serves to advantage particular actors.

In some cases, this may extend to forms of regulatory capture, where complexity disproportionately benefits those best positioned to navigate or shape it.

Implications

Policymakers introduce more:

conditions
reporting requirements
fine-grained rules

Testable Results

If this dynamic holds, we would expect:

more numerous and specific conditions attached to policies and programs;
growth in reporting requirements and data collection;
increased use of fine-grained rules and conditional logic;
a gap between nominal compliance burden and underlying structural complexity; and
differential impact, with better-resourced actors more able to navigate requirements.

3. The Overton window, revisited

Social media has expanded the set of policy ideas that are visible and discussable within political systems. Ideas that would previously have remained marginal can now be articulated, circulated, and supported by distributed groups.

AI extends this dynamic by reducing the cost of expressing these ideas in forms that are legible to the policy process. Proposals can be drafted, structured, and supported by analysis in ways that resemble traditional policy artefacts.

The result is not only a wider range of views, but a larger set of proposals that meet the threshold for institutional consideration.

Implications

a wider range of policy proposals treated as viable or “in scope”;
increased competition among proposals for attention within decision-making processes; and
greater difficulty in aggregating preferences across a more diverse set of options.

Testable results

a broader distribution of policy positions reflected in formal submissions and consultation processes;
increased number of distinct policy variants proposed within a single issue area; and
greater divergence between publicly discussed proposals and those ultimately adopted.

4. Policy paralysis

In more contentious policy domains, as more actors are able to participate more quickly and more extensively — across both traditional and newer venues — coordination costs are likely to rise.

Faster feedback loops do not necessarily produce faster agreement. Proposals can be generated, refined, and contested more rapidly, but the underlying processes of negotiation, compromise, and coalition formation remain constrained.

The system may become faster at generating disagreement than resolving it. In some cases, this may also enable more efficient forms of influence — including regulatory capture — as actors compete within a more crowded and dynamic policy environment.

Implications

more iteration without proportional increase in decisions
greater difficulty forming stable coalitions in contested areas
increased importance of process, positioning, and timing in shaping outcomes

Testable results

increased number of consultation rounds and policy revisions in contentious domains
longer time to reach final decisions or implement policy
greater incidence of partial or incremental outcomes in place of comprehensive reform
concentration of outcomes among actors able to engage persistently and effectively over time

What would falsify this?

One must be humble about predicting the consequences of the fast and wide-ranging technological and social change we are experiencing. This analytical model would be weakened if we observe:

decision speed increases significantly
agenda-setting power diffuses rather than concentrates
lobbying shifts toward mass participation over access
regulatory complexity stabilises or declines

These effects are unlikely to be uniform across policy domains, but the underlying shift in constraint is likely to be general.

Closing (100–150 words)

AI does not remove constraints from the policy process. It changes where they bind.

The cost of producing policy — drafting, modelling, and generating proposals — is falling rapidly. The capacity to evaluate, coordinate, and decide is not. As a result, the constraint shifts from production to attention and coordination.

This has several implications. Influence is likely to attach more strongly to those who shape what is considered, rather than those who produce analysis. Policy may become more complex, as the cost of managing that complexity falls. And in more contested domains, faster iteration may not lead to faster agreement.

These are not predictions of dysfunction. They are predictions of a system adjusting to a new set of constraints.

The central question is not whether AI will make policy-making more capable, but how institutions adapt to a world in which policy is abundant and attention is scarce.