October 3, 2026
David Robinson, a safety employee at OpenAI, resigned this week and told TechCrunch the company's culture is "broken." He also admitted, pretty candidly, that he's something of a cliché at this point: an AI-lab employee who leaves and immediately sounds the alarm on the way out. That self-awareness is almost the most interesting part of the story. It's not the first safety-related exit from a frontier lab this year, and it won't be the last.
Here's why a business reader should care, even if you don't work in AI safety research: every one of these departures is a data point about how the companies building the tools you're adopting actually operate internally. If the people whose entire job is to flag risk keep walking out the door, that's not just an HR story -- it's a signal about how much weight safety concerns actually carry versus shipping speed inside these labs. You don't have to take Robinson's word as gospel (he has an obvious incentive to frame his exit as principled, and OpenAI would tell a different version of events), but a pattern of resignations like this is harder to wave away than any single one.
For companies evaluating whether to build critical workflows on top of a given AI vendor, this is a reminder that vendor diligence isn't just about uptime and pricing. It's also about whether the vendor's internal incentives are pointed toward responsible deployment or toward being first. That's part of why we think about security and trust posture as a first-class evaluation criterion, not an afterthought -- see our own Trust Center for how we think about that obligation on our side of the table.
At TechCrunch Disrupt 2026, Clay co-founder and CEO Kareem Amin is set to talk about the rise of the "GTM engineer" -- a role sitting somewhere between sales ops, data engineering, and automation building, built around orchestrating AI agents to run outbound and pipeline work. It's a small headline on its own, but it's worth paying attention to because it's a symptom of a bigger shift: go-to-market work is being re-architected around people who can wire together AI tools, not just people who can write a good email sequence.
My honest take: this is both a real opportunity and a trap. The opportunity is obvious -- teams that figure out how to let agents handle research, enrichment, and sequencing can move faster with fewer bodies. The trap is that "GTM engineer" can become code for quietly outsourcing judgment calls to a black box, with nobody clearly responsible when the automation gets something wrong -- a bad outreach cadence, a compliance miss, a tone-deaf message sent at scale. The same accountability question from the OpenAI story shows up here in miniature: when a role gets built around orchestrating AI rather than doing the work directly, somebody still has to own what the AI produces. Teams building out this kind of automation internally should think hard about where human review sits in that pipeline, which is exactly the kind of thing we cover in our guide to workflow automation and in how we've built out marketing and campaign tooling and CRM workflows that keep a human in the loop rather than removing them entirely.
Put the Robinson resignation next to the GTM engineer trend and you get a pretty clear through-line: as AI gets embedded deeper into both the labs building it and the companies deploying it, the question of who is accountable for outcomes keeps getting harder to answer and more important to ask. That's not a reason to slow-walk adoption. It is a reason to be deliberate about where you put guardrails, who signs off on agent-driven decisions, and which vendors you're willing to trust with judgment calls versus just execution.
So here's the question for you: when your team hands a workflow over to an AI agent -- whether it's a safety-critical decision at a frontier lab or a sales sequence run by a "GTM engineer" -- who actually signs off before it goes out the door, and would you be comfortable explaining that answer to a customer?
Sources