News You Can Use

Revisiting the 'jagged frontier' argument for AI. Stating that AI's progress is uneven - with some capabilities far ahead and others lagging. Even as AI becomes superhuman on complex tasks like reasoning or reading large datasets, it still struggles with simpler but critical aspects like memory, edge-case handling, or interacting with the real world. Mollick explains that AI's overall ability frontier remains jagged, meaning AI will complement humans for the foreseeable future; and that breakthroughs happen not continuously, but when specific bottlenecks are overcome.

This perspective helps temper expectations about what AI can reliably do in practice. It reinforces the idea that AI will augment legal work by accelerating parts of tasks where it excels, but will still require human intervention where it struggles (e.g., retaining context over long matters or handling unusual/rare exceptions). When evaluating and deploying AI tools, whether internal pilots or client solutions, we should focus on where AI boosts productivity most and understand the limitations that will persist until key bottlenecks are addressed. Just because AI can do one thing well, does not necessarily mean it will do other things of the same 'complexity' as seen by humans.

This article highlights how AI tools are transforming traditional legal due diligence. With M&A deal value rising sharply, due diligence has historically been expensive and slow, often costing 0.5-2% of deal value and taking weeks or months. AI-powered platforms can now scan documents, extract key clauses, flag risks, and surface inconsistencies much faster than humans, helping teams complete reviews in days rather than weeks. The core shift is that AI handles the first-pass review, while lawyers focus on judgement, negotiation, and deal strategy.

This is a good example of AI delivering real efficiency gains in a well-defined legal workflow, rather than just generic productivity uplift. It reinforces where AI makes the most sense today: high-volume, time-critical exercises where speed and consistency matter, and where lawyers' time is better spent advising on risk rather than extracting data. It also sets a useful benchmark when evaluating tools - the question is whether it meaningfully reduces turnaround time, cost, or pressure on teams during live deals. We are already seeing some of our Private Equity clients investing in automated DD and self-serving these projects.

The author argues that venture incentives prioritise rapid valuation growth and distribution over truly robust legal products - many legal AI offerings end up resembling repackaged general-purpose models with little real differentiation. Lawyers' need for high reliability, perfect formatting, data security and deep domain trust makes legal AI a tough nut for VC-backed startups to crack.

This helps explain the current market dynamics: even with heavy investment, legal AI solutions struggle to convince lawyers they are reliable enough for daily use. This means we should be cautious about equating funding buzz with product quality. When assessing tools for internal or client use, we should prioritise trustworthiness, accuracy, and workflow fit over hype or headline valuations - and clearly communicate where products truly add value versus where they rebrand generic capabilities. It also matches up with our frustration in the market, distribution is more important for VC than the development which means the products end up being too shallow in areas we need more in depth capabilities.