BairesDev's survey finds four in five technology leaders pressured to overstate AI progress

The loop De Marco describes runs through the board pack. Leadership sets aggressive targets, teams present a picture rosier than reality supports, and leadership makes the next round of decisions on that data ^[16]. In his study of more than 500 technology leaders, 79% said they feel pressure to overstate AI progress, and nearly half placed the origin of that pressure in the C-suite or board ^[7][8]. If nearly half of that 79% named the board or the C-suite, the share of the whole sample pointing upward is about 38% ^[21].

In a separate BairesDev survey of more than 1,300 developers across 61 countries, 10% said accountability for AI outcomes sits with senior leadership and over half said it sits with them personally ^[5][6]. Developers naming themselves outnumber those naming the top by at least five to one ^[20]. De Marco wrote that "it just never gets named, and by the time anyone notices, the team closest to the work has already absorbed it" ^[23].

The readiness estimate is the softest figure in the piece and the one the strategy rests on. A fellow at BairesDev puts roughly 70% of what AI success requires in people and process change, 20% in tooling and 10% in the models themselves, and De Marco wrote that budget and attention flow in the opposite direction ^[11][12]. Tooling and models together come to 30% of that estimate, so the two lines a board can approve in a single meeting cover less than a third of what this framing says decides the outcome ^[19]. It is one colleague's estimate, not a measurement.

Where that commitment gets expensive is institutional knowledge: much of what the systems depend on lives in people's heads and not in documentation or data pipelines, De Marco wrote ^[17]. His description of the failure: "A model operating against incomplete context produces confident output that sails through review because the reviewer is missing the same context the system is" ^[14].

The conflict is in the byline. A company that sells embedded software teams to more than 500 clients has an interest in telling boards that their AI problem is people and process, because people and process are what it sells ^[1][2]. Two of the findings cut against the convenience. One puts the disruption inside the boardroom: 88% of technology decision-makers said at least one active AI initiative was significantly disrupted by shifting executive priorities in the previous year ^[10]. The other puts the binding constraint on quality in the schedule, with deadlines and delivery pressure reported as the top barriers to validating AI output ^[9].

So this quarter's decision is narrow. A board approving an AI target can name who owns the outcome and how much validation time the schedule contains, or it can leave both unstated and read next quarter's progress report knowing what produced the numbers in it. De Marco's line on the second path: "Leaders who compress timelines without adjusting quality expectations are making a quality decision by default rather than by design" ^[13]. None of the three studies tests whether programs with a named accountable owner delivered better than programs without one.