EthosPowerAIA tender arrives on a Tuesday with the client's own IEEE 485 battery sizing clipped to it, the schematics and the bill of materials produced overnight by a language model that read four datasheets and an old copy of the standard, and the numbers are not obviously wrong. The customer changed, not the technology. Whatever margin rested on the client not knowing how to read a standard is closing, and it will not reopen; what remains worth paying for is the judgement no datasheet contains.
Sell the judgement
The services that were the product are becoming the free part. Sizing, a first-pass schematic, a compliant parts list: a customer with a model and an afternoon now produces these to a standard that used to take a junior engineer a week, so the firm that sold the week has to sell something else. What no datasheet contains is how installations actually degrade, which network operator is enforcing which edition of AS/NZS 4777.2 this quarter, and whether a UL 9540A report covers the configuration being sold rather than the one that was tested. That judgement was always the valuable part, and it is now the only part left with a price on it. Sell that.
AI aimed at the low-hanging fruit works for a while and then drains productivity, because a tool dropped into an unchanged process makes the process faster at producing what the customer no longer needs. The better use of a firm's data is to turn it outward: to act as quality assurance for the client's own supply chain, to monitor the installed base the firm already sold and advise on it, to be in the room before the specification is fixed rather than after the shortlist is.
Let the business owners build the prototype
The people who know what the business sells can now build the first version of the tool that sells it, and that removes a translation step that has cost projects for as long as software has existed. A manager who sketches the workflow with a model, watches it fail on the real edge cases and fixes the logic before an engineer is involved has specified the system in the only language that matters. Technology remains the third and minor part of the change, after the business model and the organisation, but it is the part that should be played with early and often, because the prototypes are how a leader finds out what the industry will look like in five years rather than what the tool can do this afternoon.
Screen the expert
A firm hiring someone to lead this should screen hard, and Maxim Afanasyev's criteria are the ones to use. The first is that the knowledge cannot be bought quickly: the candidate who came to AI from ESG or blockchain after a two-week course, because the last field cooled, is not the candidate, and the ones worth having show years of accumulated work rather than a recent pivot. The second is that three capabilities have to sit in one head, knowing how the business makes money, the transformation skill to redesign a workflow and win adoption for it, and enough technology to make sound architectural choices; three specialists in a room do not substitute, because the interconnections are the job. The third is that the role reports to the chief executive or sits on the board with authority over IT and operations, since structural change cannot be executed from the middle. Nor from the side. The fourth is that the candidate treats the technology as the minor part and the redesign of what the firm sells as the work, which Afanasyev calls mechanism design. Code is the easy part. The fifth is that, because such people are scarce, the good one builds an internal incubator to grow the next ones, led by strategic business thinkers looking five years out rather than by an engineering team testing tools.
The argument is Afanasyev's, Financial Services Industry Head for Asia Pacific and Japan at Google Cloud, and it is one practitioner's experience with no sample and no control; Google, like McKinsey, sells the remedy. His interviews, the phrases quoted here, and what I take them to mean for battery-backed power are set out on the approach page.