AI is dissolving the line between engineers and everyone else, and org charts are already changing to reflect it
The assumption that engineers write code and everyone else waits is breaking. From Anthropic's internal teams to Instagram's restructured pods to a company that canceled a $600,000 Salesforce contract, the evidence points in one direction.
The organizational assumption that has governed software product development for decades is simple: engineers write code, and everyone else waits. That assumption is breaking. The evidence comes from teams across the industry, and it points in one direction.
Fiona Fung, describing the Clockwork team at Anthropic, reports that designers and PMs now check in code alongside engineers. The bottleneck that once forced a PM with a clear idea to queue behind engineering bandwidth has largely disappeared. That is not a cultural shift or a management philosophy. It is a change in what the tools allow.
Andrew Ambrosino’s observation from OpenAI adds a telling data point about where non-engineers actually go when given the choice. When Codex, a developer tool, was available alongside products built specifically for non-developer personas, non-developers chose Codex. As Ambrosino puts it, “nobody would leave the Codex app for the apps that were allegedly for these other personas.” The implication is that the most useful thing a non-engineer can do with AI is not to use a simplified tool designed for them. It is to reach directly for the tool that does the most.
We just recently canceled our Salesforce contract because we have an internal CRM that was built, you know, was vibe coded, that is working better, that is managing our process better, is more integrated into what we're doing. We run our agents inside of it and no one was using Salesforce anymore. $600,000 a year. Gone to zero. Fred Turner
Instagram is formalizing this shift at the structural level. Adam Mosseri describes a move away from canonical teams of around 13 people toward smaller pods of four to six engineers paired with a new role called “product staff.” The product staff role is an evolution of the PM, one that can perform tasks traditionally handled by designers, data scientists, and researchers, using internal AI tools to do so. Mosseri says this year the change is taking hold. The org chart is being rewritten not to accommodate a theory but to reflect what the tools already make possible.
The business-impact dimension is clearest in Fred Turner’s account of Curative. The company replaced its Salesforce contract, at $600,000 per year, with an internally built CRM that was vibe-coded and integrates AI agents directly. Turner describes the internal tool as working better and more integrated into the company’s actual processes. The dollar figure is striking, but the more consequential detail is what Turner says came next. Rather than using AI agents to maintain the same contract volume with fewer people, Curative used the unlocked capacity to attempt ten times as many contracts annually as it could the prior year. Removing the engineering bottleneck did not just make the existing operation cheaper. It changed the scale of what the operation could attempt.
Aaron Levie adds a note of structural continuity worth holding onto. Vibe coding and AI-assisted development, in his view, will not simply erase existing enterprise software. The more likely path is that these tools get layered on top of data stacks that organizations already trust, with IT personnel and non-technical staff customizing workflows at a level of specificity that was previously impractical. That framing does not contradict the pattern; it describes the mechanism. The boundary between technical and non-technical work is not disappearing because existing systems are being thrown out. It is dissolving because the cost of building on top of them has collapsed.
What connects these accounts is not just speed or efficiency. It is a shift in who can make things. The filters that separated builders from specifiers, engineers from everyone else, were partly a function of skill and partly a function of tool access. Both are changing simultaneously. The org charts, the role definitions, the headcount ratios, and the planning assumptions most product organizations run on were calibrated for a world where writing code was expensive and slow. That world is receding faster than most structures built around it.