They say AI isn’t here to replace engineers, it’s here to augment them. That’s the polite version. What surfaced at ELC’s July 10 roundtable was something else entirely. This wasn’t your usual talking-heads panel. This was a wide-open signal flare across engineering leadership: the AI era isn’t coming, it’s already refactoring how teams ship, scale, and structure.
When Santha Nandakumar hosts a session, the energy shifts. She doesn’t just facilitate, she calibrates the room. Decades of engineering leadership at Capital One, Movius, PRGX, and Lavante give her the range. Now, as an executive coach at Tandem, she’s helping engineering leaders do what most of these tools can’t: navigate the human side of velocity. And that’s exactly what this roundtable delivered.
This wasn’t a GitHub Copilot fan club meeting. It was a front-row look at what happens after AI tools land inside the org. Copilot’s market share? 42%. Adoption? 90% of engineering teams. PR cycle times? Now 1.2x faster with AI. But those stats don’t tell the whole story. The real headline is in the complexity beneath the adoption curve. Different teams. Different tools. Different results. The ground truth isn’t clean. It’s layered.
Inside Intuit, Manoj Mohan wasn’t focused on how to “do something with AI.” That framing, he argued, is where most teams go wrong. Start with pain, not hype. At Intuit, that meant reducing 90-minute incident response times across 50+ microservices. Their solution? An internal AI agent built on prompt engineering that surfaced probable causes via Slack, trained on Splunk logs, Wavefront traces, and runbooks. The system took on 60% of the typical L1-to-L2 workload. It didn’t replace anyone. It just made their lives less hellish during production calls.
At Equinix, DT rolled out a RAG-powered agent to unify customer support signals buried in Salesforce, Jira, ServiceNow, and Slack. Suddenly, support escalations started arriving with actual context. L1 could handle more. Developers could focus. Mean time to resolve dropped by 40%. Productivity didn’t just improve, it got smarter.
But for many in the room, the story was less polished. AI tools are everywhere, Copilot, Claude, Cursor, Gemini, Zoom Companions, Perplexity, and the signal-to-noise ratio is brutal. Tim Delesio at Doxo called out the fragmentation. Different engineers using different tools, building parallel workflows. Custom agents. Conflicting preferences. No clear standard. Tooling without integration is just overhead with good branding.
Still, the tools are sticking. GitHub Copilot isn’t just paying for itself. Teams are modeling ROI against the cost of not using it. As one attendee put it, “If an AI agent costs $30K a year and delivers a 10% gain, it pays for itself before the second sprint.” And if that sounds theoretical, it isn’t. PRs with AI saw a 260% year-over-year usage jump, from 14% to 51%. Engineering orgs are now comparing the cost of agents versus offshore devs when making hiring decisions. These are board-level tradeoffs, not experiments.
The effects are structural. Flat hierarchies. Fewer backfills. PMs are using Claude to generate full PRDs. Engineers are building internal copilots for CLI workflows. Everyone’s asking: where’s the value, and who actually owns it? At one point, the group acknowledged that developers are now building PRDs from Zoom transcripts while PMs are prototyping with agents. The lines aren’t just blurring, they’re folding into each other.
And that’s where leadership hits the spotlight. Teams are asking for upskilling support, but most orgs are still treating AI fluency like optional training. Those leading the curve are treating AI like a new dialect, everyone speaks it differently, but alignment only happens when the whole room can hold a conversation. Whether through AWS workshops, Claude-based experimentation, or weekly “developer congress” demos, the message is clear: you don’t just adopt AI, you socialize it.
When the conversation turned to compliance, the room tightened up. GDPR. PII. Multi-tenancy. Role-based access. The AI agents are in production now. They’re not vaporware. Which means they need access credentials, guardrails, and governance just like your engineers do. As Kusumita noted in chat, “AI agents are quietly becoming the shadow users of the enterprise.” If you’re not managing them like real users, you’re already behind.
So what does a high-performing engineering team look like two years from now? One answer landed like a headline: humans and agents, side by side. Not because the humans couldn’t handle the work, but because doing it without augmentation would be inefficient. Another attendee (Full disclosure – Me 😉 ) put it sharper: “The machine used to tell us when we were wrong. Now we have to tell the machine when it’s wrong.” The power shift is happening. But only the orgs who embrace that mental model will know how to wield it.
Over 10,000 engineering leaders are now part of the ELC community, across more than 3,000 companies. These roundtables aren’t about trends. They’re about readiness. At 90% AI adoption, the conversation isn’t if you’ll integrate. It’s whether your team culture, your decision frameworks, and your leadership muscle are ready to scale with it.
AI isn’t a sidebar to engineering anymore. It’s in the loop. In the sprint. In the staffing models. In the Slack threads. And if you’re still treating it like an optional upgrade, don’t be surprised when your top performers start working for someone who doesn’t.
If engineering peace of mind is what you crave, Vention is your zen.
