Why AI Will Accelerate Bad Development Decisions Before It Improves Good Ones

What developers, investors and planners still need human commercial judgement to do in an LLM-driven market

AI is moving into property faster than expected.

Feasibility summaries can now be produced in seconds. Planning policy is scanned instantly. Cost benchmarks, market commentary and risk registers arrive fully formed, written with a compelling level of confidence. For developers, investors and advisers operating under time pressure, this might feel like progress.

The question is no longer whether AI will change how decisions are made in development, because it already has. The big question is what happens when the tools shaping those decisions do not reflect the realities of development risk.

Across the UK market, the most consequential decisions are being made earlier than ever, often before a design team is properly established or consultants have begun to interrogate assumptions. They are made before costs have hardened and before constraints have been fully tested.

A combination of safeguarding constraints, infrastructure thresholds, political context, aviation restrictions, utilities capacity, Section 106 exposure and market timing is now shaping outcomes well in advance of planning committee. On more complex schemes, this dynamic is already well understood.

What has changed is that AI now sits upstream of these conversations.

Large language models are effective at organising existing information. They can easily summarise policy, reflect precedent and construct a coherent narrative. However, they lack the ability to judge which risks are material and which are merely documented.

In development, failure seldom stems from a shortage of information. More often, it arises where risks are misread, underplayed or not fully understood in the context of delivery. This is where the limitations of AI become evident.

Planning frameworks highlight the issue. Policy that appears flexible can prove restrictive, safeguarding that looks workable can undermine yield, and infrastructure assumptions can unravel once timing, funding and politics come into play. AI can outline these factors, but not how they interact in reality. The risk, therefore, is not necessarily incorrect information, but misplaced confidence. AI produces convincing outputs that create a misleading sense of certainty.

Once an early narrative takes hold, it quickly becomes embedded. Cost plans align to it, design work develops around it and funding conversations begin to reflect it. By the time a senior professional challenges the premise, sunk time and reputational momentum make it difficult to change direction.

This dynamic is already visible in the market.

At the same time, conditions are less forgiving than in previous cycles. Margins are tighter, time carries a higher cost and the ability to recover from early misjudgement is significantly reduced. Decisions do not need to be reckless to undermine a scheme; they only need to be incrementally wrong.

In this context, human commercial judgement remains critical and cannot be replaced by automated output.

Strong development outcomes depend less on knowing the rules and more on understanding where they bend or conflict. They rely on delivery experience and on knowing which assumptions carry real consequence.

AI lacks this awareness. It cannot read political context, anticipate delivery friction or distinguish between risks that matter a little and those that matter a lot.

This is where experienced professionals continue to add value.

In a market shaped by widespread access to AI, value shifts from producing information to interpreting it. This means testing assumptions, challenging narratives and applying judgement where logic alone is insufficient, particularly at the earliest stages where small decisions have long-term impact.

This does not diminish the value of AI. Used well, it can accelerate analysis, broaden options and remove friction from early work. It remains a powerful tool within the process, but it does not replace decision-making responsibility.