Due diligence exists because the people buying expertise cannot independently evaluate it. The frameworks that have developed around this problem — credential verification, portfolio review, reference checks, structured interviews, work samples — were built on a single foundational assumption: that producing expert-quality output required expert-level competence. The evaluation frameworks were proxies for competence because competence was what produced the outputs being evaluated.
That assumption is no longer reliable. The frameworks built on it are failing silently, and the organizations using them do not know it yet.
What Due Diligence Was Designed to Detect
The purpose of professional due diligence is not to confirm that a candidate or vendor has produced good work. It is to establish a reasonable basis for believing they can produce good work under the specific conditions of the engagement being considered.
This distinction matters. Past outputs establish that someone produced something. They do not establish that the person who produced it understood it, can replicate it under different conditions, or can adapt when the conditions diverge from expectations.
Traditional due diligence frameworks addressed this gap imperfectly but adequately for most of professional history. Credentials established that someone had completed a recognized program of study. Portfolio work demonstrated outputs across multiple contexts. References confirmed that previous clients had been satisfied. Structured interviews probed reasoning and judgment. Together, these inputs created a picture that was correlated, if imperfectly, with actual competence.
The correlation held because the barrier to producing impressive credentials, portfolio work, and articulate responses in an interview was, in most cases, the underlying competence those things were supposed to represent. Faking competence at the level required to pass rigorous due diligence was difficult. It required sustained investment in developing the surface signals of expertise, which typically meant developing some portion of the underlying expertise itself.
AI has broken this correlation. The investment required to produce surface signals of expertise has collapsed. The underlying competence those signals were supposed to represent has not become easier to acquire.
Where Each Framework Fails
Credential verification confirms that someone completed a program. It does not confirm that they retained the knowledge, applied it in practice, or developed the judgment that comes from working in the field over time. Credentials were always a weak signal. They are now weaker, because the coursework that produces them is increasingly completable with AI assistance, and because the credential itself says nothing about how the person performs in unstructured real-world conditions.
Portfolio review is more vulnerable than credentials because the outputs being evaluated are exactly what AI produces well. A portfolio of market analyses, strategic plans, compliance frameworks, technical reports, or research summaries can now be produced to a high surface standard by someone who cannot explain the reasoning behind any of it. The reviewer evaluating the portfolio is typically looking at whether the work is coherent, comprehensive, and professionally presented. It will be all three. The question of whether the person who produced it understands it is not answered by looking at the output.
Reference checks have always been the weakest element of due diligence because references are selected by the person being evaluated. They are now weaker still because the gap between impressive deliverable and actual competence can persist through multiple engagements before it becomes visible. A client who received polished work and experienced no visible failure has no basis for a negative reference, regardless of whether the work was substantively correct.
Structured interviews probe reasoning and judgment. This is where AI-era due diligence should concentrate. It is also where it most consistently fails. Interviewers ask questions that can be prepared for. Candidates prepare. The preparation increasingly involves AI-assisted rehearsal that produces fluent, well-structured answers to anticipated questions. The ability to produce a fluent answer to a prepared question is not the same as the ability to reason through an unanticipated problem under time pressure with real consequences.
The Confidence Gap
There is a specific failure mode that AI has introduced into the due diligence process that did not exist at meaningful scale before.
In most professional domains, genuine expertise is accompanied by a specific kind of calibrated uncertainty. The expert knows what they know and knows what they do not know. Their confidence is proportional to their understanding. They hedge in the right places, flag the limits of their analysis, and identify the conditions under which their recommendations might not hold.
AI-assisted non-expertise tends to produce the opposite pattern. The output is confident where it should hedge, comprehensive where it should acknowledge limits, and fluent in exactly the places where genuine uncertainty would appear in the work of someone who understood the problem deeply.
This inverted confidence pattern is a signal. It is not a reliable signal, because some genuine experts are also overconfident. But it is a signal that due diligence processes are not currently designed to detect, partly because overconfidence in professional settings has historically been correlated with seniority rather than incompetence.
What Adequate Due Diligence Requires Now
The frameworks that are adequate to the current environment share a common feature: they test process, not output.
Testing process means asking not what someone produced but how they produced it. What was the analytical approach? Where did the initial framing prove inadequate and how was it revised? What alternatives were considered and rejected, and on what basis? What would change the conclusion, and under what conditions would the recommendation not hold?
These questions cannot be rehearsed effectively without the underlying competence to answer them, because they require navigating ambiguity in real time rather than retrieving prepared responses. An AI-assisted non-expert who can produce a polished market entry analysis cannot explain why the three alternatives they considered were inferior without either revealing that they did not consider alternatives or producing answers that reveal the absence of genuine analytical judgment.
This approach requires more from evaluators. It requires domain knowledge sufficient to assess the quality of reasoning, not just the quality of presentation. Organizations that lack that internal expertise face a structural problem: they cannot evaluate what they are buying. The solution is either to develop that internal expertise or to engage someone who has it specifically to conduct the evaluation.
Neither is convenient. Both are necessary.
The Market Consequence
Due diligence failures at scale produce a specific market outcome. When buyers cannot reliably distinguish between genuine expertise and AI-assisted surface competence, the market price of expertise declines toward the price of the surface signals. Genuine experts face price pressure from competitors who can produce equivalent-looking outputs at lower cost. The quality of professional services in the market deteriorates while the appearance of quality is maintained.
This is not a prediction. It is an observable dynamic in professional services markets where the evaluation frameworks have not kept pace with what AI has made possible. The organizations experiencing it are those that optimized their procurement processes for cost and efficiency rather than for the ability to evaluate what they were actually buying.
The correction will come from the engagements that fail. It will come from the due diligence findings that do not hold up. It will come from the regulatory examinations that find documentation without operations behind it. It will come, in other words, from the point of consequence — which is exactly where the gap between apparent and actual competence always becomes visible.
The question for organizations now is whether they wait for that correction or build the evaluation capacity to avoid it.