Nacho De Marco is the CEO of BairesDev, an award-winning nearshore software outsourcing company, and the cofounder of VC firm BDev Ventures.
getty
Over the past two years, companies have moved from unstructured AI experimentation to embedding AI across entire business functions—faster, in many cases, than the organizational thinking required to support it. As with every major technology cycle, the gaps reveal themselves after wide adoption. One of those gaps is accountability for AI outputs. When AI-assisted work produces a failure, a wrong decision or an output no one can stand behind, the question of who is responsible tends to land on the developer.
Most organizations don't realize they designed it that way, intentionally or not. Developers are taking on accountability by default, through individual judgment calls and self-imposed workarounds, where formal guidance is limited or absent.
We have data on this and daily visibility into how it plays out inside real teams, across real organizations and at scale.
In Q1 2026, BairesDev surveyed 1,329 developers across 61 countries. Over half (51%) say accountability for AI-driven outcomes sits with them individually. Only 19% see it as a team responsibility.
I think two forces explain this, and they're worth separating.
The first is organizational. Most organizations are moving faster than their governance can follow. AI capabilities outpace internal assessment, and because of that, the frameworks meant to manage that risk tend to be layered on top of existing workflows.
Executives are prioritizing AI as an investment, but accountability rarely makes it into that conversation. The result is a gradual erosion. Delivery pressure builds, human judgment leaves the process incrementally and what fills the gap is the assumption that moving fast and following steps is the same as getting it right.
The second force is structural. AI-generated code is probabilistic and pattern-based. It can produce outputs that look credible and well-formed, and still be wrong in ways that only surface when someone closer to the domain takes a hard look. Every consequential output requires validation from someone, or something, with the context to judge it—whether that's technical, operational or both. Most organizations, from what we observe, haven't fully answered that question yet.
Because BairesDev developers work embedded inside client teams across industries, from growth-stage startups to Fortune 500 enterprises, we see governance structures at scale.
Across hundreds of simultaneous projects, the pattern we observe is consistent: Governance structures are still evolving across the industry. As a result, the question of accountability doesn't always have a clear, shared answer, and in practice, it often settles with the individual closest to the work, expanding the scope of what they carry.
What we observe and what senior engineers on our teams confirm is that accountability tends to default to whoever has the deepest familiarity with the code and understands the implications beyond it. They see the impact on business processes, stakeholders and decisions that were never purely technical to begin with.
In many cases, a developer may be the only one examining the technical layer, even when the consequences extend well beyond it. And regardless of what governance structure exists around them, they navigate it and fill the gaps where it doesn't.
In my experience, what closes the gap is making accountability explicit and operational—not more documentation, but clearer design. Here is what that actually requires:
1. Design accountability into the workflow, not around it. Clear decision rights over AI-assisted outputs need to be built into how work flows, not assumed to exist because reviewers are assigned or policies are documented. Documentation doesn't hold under delivery pressure, and the ownership that isn't enforced isn't ownership.
2. Assign responsibility to the right person. Someone needs to own AI-assisted outcomes with the authority and context to understand their impact beyond technical correctness. That can't default to seniority alone. The title doesn't confer the institutional knowledge needed to understand what a decision costs across business processes and stakeholder relationships.
3. Reset expectations about what AI outputs actually require. When AI outputs are treated as reliable by default, accountability silently shifts to the developer without the support structures to match. A common misconception that continues to show up is the assumption that lower effort from AI means lower effort overall. In practice, it doesn’t. It shifts the effort to validation, and if that validation isn’t clearly designed into the workflow —whether through people or systems—it lands on whoever is closest to the output.
4. Apply the same governance discipline to AI that you apply to code. Version prompts and workflows so drift is traceable. Apply pre-production standards to AI outputs before they reach production. These aren't exotic practices; they're the same principles engineering teams already apply to software. The gap, as I see it, is that leaders haven't yet decided that they apply to AI, too.
Expecting someone to manually review every AI-generated output doesn’t scale. As AI systems evolve, validation shifts into infrastructure: evals, monitoring and guardrails. What doesn’t change is accountability for outcomes and where that accountability lives.
Developers are ready. The tools are moving fast. The gap is organizational.
AI accountability doesn't emerge naturally from good engineering or better models. It has to be designed: assigned, supported and revisited as the technology changes. Left undefined, accountability settles, by default, on the individuals who know the code best but don't own the business outcomes it affects.
How accountability gets defined from here is a leadership decision. The question for every tech leader reading this is whether their organization has made it yet.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。