Leadership1 publisher2 min readPublished
Lugo: Developers should be judged on problem-solving with AI, not just coding speed
INVID's Alberto Lugo argues in Forbes that developers will be judged on how accurately they solve problems, and his case rests on a Gartner adoption forecast and a McKinsey productivity finding, both cited secondhand.
The Board Room · Leadership desk
What happened
- He grounds that prediction in a Gartner estimate that 90% of enterprise engineers will have AI coding assistants by 2028.
- He also cites a McKinsey report finding that developers can complete coding tasks in half the time when they use generative AI tools.
- The same column argues technical expertise becomes more important, listing architecture, cybersecurity, data management, cloud infrastructure and system integration as what lets a developer spot wrong AI output.
- Lugo states plainly that AI is not eliminating the developer role, since humans are still needed to support code writing, fix bugs and upgrade software.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- decision Twice as many completed coding tasks per hour puts review capacity on this quarter's agenda ahead of ladder language, because the reading has to be staffed before the criteria are rewritten.
- constraint A manager rewriting criteria has to decide what a verification-weighted ladder stops rewarding, and that subtraction is where the internal cost of the change sits.
- exposure If accountability for the finished product stays with the developer, the engineer approving generated code is the one answerable when it fails in production.
- cost A mid-cycle rubric rewrite is paid by the engineers already being graded, whose completed year of work gets revalued against criteria written for conditions expected in 2028.
A forecast about what sits on an engineer's desk counts equipment. Gartner's 2028 estimate, as Lugo cites it, describes how many engineers will have an assistant [1]. A promotion rubric has to score a behaviour, which means picking something countable, such as the rate at which bad generated code gets past review. What the column offers instead is a list of capabilities: prompting AI with relevant and highly honed questions, validating outputs, understanding the limitations of AI models, and maintaining accountability for the final product [8]. Those are plausible criteria, and none of them is stated as something a manager can count.
The McKinsey finding Lugo cites carries the nearer consequence. If a coding task takes half the time, the same engineer finishes roughly twice as many of those tasks per hour [1]. Somebody still has to read the output, and under Lugo's own framing that somebody is human [8].
The prediction at the centre of the column is about evaluation. "The software developer of the future will be evaluated not by how quickly they can write lines of good code," Lugo wrote, "but how accurately they can solve problems and leverage the power of AI to enable real-world outcomes" [3]. He is the founder and president of INVID, a Puerto Rico-based software and AI solutions provider with customers across the United States [2]. The piece makes the case for that standard; it does not cite a company that has rewritten its ladder.
Senior ladders already reward scope and design judgement, and a 2028 adoption number changes nothing about how to score this year's cycle. On timing, that reading holds. Under the same Gartner estimate, one enterprise engineer in ten is still working without an assistant in 2028 [2], so a rubric written as though the tool were universal is pricing a condition that has not arrived. Lugo's stronger claim is the one to test. He treats effective AI use as a prerequisite to the role rather than an accelerant [6], and he argues that AI does not inherently understand why a business problem matters, what a customer actually needs, or what an organisation should prioritise [7].
The two horizons pull in different directions for a manager writing criteria this quarter. Verification load is a staffing question now. Ladder language is a 2028 question. And the requirement list in the column grows at both ends. Lugo says technical depth matters more, not less: a developer needs software architecture, cybersecurity, data management, cloud infrastructure and system integration to recognise when an AI-generated solution is wrong [9]. He also tells rising developers to "think like a CEO more than a CIO" [10].
What to watch
- Gartner publishing the survey behind the 90% estimate would turn a forecast into a baseline managers can plan headcount against.
- A named employer publishing an engineering ladder with scored verification criteria would show what the behaviour looks like once it is measured.
- Defect escape rates for assistant-generated code would test whether the halved task time survives review.