AI Rewired Software Development. Cybersecurity Is Next, But Not in the Same Way

Key points

• A March 2026 Federal Reserve Board working paper found that annual coder employment growth is roughly 3% lower than it was before ChatGPT launched.
• Google Cloud's 2025 State of AI-Assisted Software Development report found that 90% of developers now use AI as part of their daily work.
• Gartner predicts that 80% of organisations will shrink large software engineering teams into smaller, AI-assisted teams by 2030.
• AI agents are already handling first-level alert triage inside security operations centers, with humans managing escalation and high-stakes decisions.
• Caleb Sima, chair of the Cloud Security Alliance AI Safety Initiative, says fully autonomous vulnerability testing in live corporate networks is still some way off.

Software developers saw it happen fast. A job that once meant a single person writing code alone for hours has become something closer to conducting a chorus of AI assistants. Ninety percent of developers now use AI tools at work, according to Google Cloud's 2025 research, and developers who use AI complete 26% more tasks than those who don't, per an earlier Microsoft study. The Federal Reserve Board noted in a March 2026 working paper that the flip side is real: annual coder job growth is now about 3 percentage points lower than it was before AI chatbots arrived.

Security researchers and executives watching that shift are now asking whether their field comes next.

How close is an AI-run security operations center?

Pretty close, actually. A security operations center, or SOC, is the team inside a company that watches for cyberattacks around the clock and decides which alerts are real threats worth acting on. That first layer of work, sorting genuine alarms from noise, is exactly what AI agents are already taking over.

Contrast Security's CISO David Lindner described a real example: an incident ticket arrived through Jira (a project-management tool), and an AI agent automatically pulled supporting information from GitHub (a code repository) and Datadog (a system-monitoring service), then delivered an initial assessment. "I don't want to even call it a junior SOC analyst," Lindner said. "It is a SOC analyst that does some initial triage."

Not everyone is ready to hand agents the final call. Lionel Litty, CISO at Menlo Security, said his team uses AI agents to provide context and prioritise alerts, but humans still decide what to act on. The middle ground most practitioners land on: agents handle first-level sorting, senior humans handle judgment and escalation. We covered the difficulty of securing the agent platforms that make this possible in our September 2026 look at the OpenClaw rebuild.

What about the flood of newly discovered security flaws?

AI is surfacing security vulnerabilities, meaning weaknesses in software that attackers can exploit, faster than most teams can fix them. That sounds useful. It mostly creates a different problem.

Caleb Sima, founding partner of White Rabbit and chair of the Cloud Security Alliance AI Safety Initiative, puts a fine point on where AI vulnerability-hunting actually works today versus where it falls short. Finding flaws in source code? He says that problem is essentially solved. Testing those flaws against a live, complicated corporate network in a fully autonomous way? Still a genuine challenge.

Lindner's summary is hard to argue with: "We don't have a problem finding problems. We have a problem triaging and remediating all the problems that we find."

Area	Where AI stands today
Sorting security alerts (first-level triage)	Agents doing this now at many organisations
Finding flaws in source code	Largely solved, per experts
Autonomous testing of live production networks	Still developing
Machine-speed attack containment	Early-stage; seen as necessary
Full SOC replacement by agents	Near-term possibility, not yet standard

Should security professionals worry about their jobs?

Not exactly, but they should expect a different job. Gartner's prediction for software teams, that 80% of organisations will replace large engineering groups with smaller, AI-assisted teams by 2030, is the template the field sees coming for security too.

Jim Reavis, CEO of the Cloud Security Alliance, told CSO Online he expects a flattening: fewer layers between leaders and hands-on builders, with senior staff directing fleets of AI agents rather than managing junior analysts. New roles, including AI governance specialists and people focused on how AI systems are instructed and overseen, will fill some of the gap. Our earlier piece on what separates a security technologist from a security executive is worth reading alongside this one.

The part that makes cybersecurity genuinely different from software development is reproducibility. A security control has to produce the same result every time; you can't ship a patch that works 80% of the time and call it good. That constraint means the field won't flip overnight the way a code editor workflow can.

Reavis put a rough timeline on it: "Two years from now, it may be unrecognizable from what it is today." That's fast. It's also not next Tuesday.

Common questions

Does any of this affect ordinary people, or just security professionals?

It affects both. Faster vulnerability discovery means software companies can patch flaws sooner, which benefits anyone who uses apps or websites. The risk is that attackers are also using AI agents to find those same flaws, so keeping software updated matters as much as ever.

What should organisations do right now to prepare?

Security practitioners recommend building environments where one breach can't bring down everything, a principle called least privilege (staff and systems only access what they genuinely need) combined with separation of duties (no single account or system controls too much). Those fundamentals hold whether the attacker is human or an AI agent.