Are we ready for so much code?

My recent conversations with customers, friends, and family about AI tend to gravitate around the fear of AI replacing our jobs. I can't confidently speak broadly about the impact to most jobs, except one that's close to home: software engineering. Recently, my posture has gone from uncertainty, to fear, to huge positivism. The more comfortable I get letting AI superpower my coding, the more I realize AI is not a replacement agent, it is an enabler.

Many non-developer roles are now coding and nobody has a plan for it

Since Q1 of 2026, I've pivoted to coding 10x more prototypes and proofs-of-concept for customers than last year. About 50% of the time I'm not throwing that code away after the demo. I'm handing it directly to customers so they can deploy it in their own environments. I've also started building my own internal tools when corporate-provided options don't exist or move too slowly. A couple of hours of vibe-coding upfront can unlock days or weeks of productivity gains down the road. Fellow CSMs I've talked to are reporting the same: 10x to 20x more code than a year ago, both for customers and for themselves.

This isn't just a CSM trend. It's showing up everywhere in my customer conversations.

Observations from the field: backlogs are getting unclogged

Customers embracing AI-enabled SDLC strategies are telling me the same story across the board:

• Tech debt that sat for years is finally getting tackled
• Deprioritized projects are back in scope
• Non-developer teams in HR, legal, finance, and marketing are shipping code with tools like Bedrock, AgentCore, and SageMaker
• Customers are buying less SaaS and building more home-grown internal tooling

The volume of code getting shipped is unlike anything I've seen in my career, and it's accelerating.

Is the infrastructure scaling?

All evidence points to… not quite, at least not for the most recognizable code host providers. GitHub is struggling to maintain reliability since the boom of AI coding. Third-party measurements put its uptime at roughly 86% recently, meaning some part of the platform has been down for an average of 2–3 hours per day, every day, for the past 90 days. In late April, a data integrity incident caused squash merges to silently drop commits from over 2,000 pull requests, with GitHub unable to help customers recover them.

GitHub's CTO attributed this to AI agent-fueled load: service traffic has grown approximately 3.5x over two years, with the steepest spike happening in recent months. GitHub only started planning for a 10x capacity increase in October 2025, and by February 2026 had already revised that target to 30x, all while mid-migration from its own data centers to Azure. In contrast, GitLab, Bitbucket, and Vercel are seeing similar load growth but not the same meltdown, raising the uncomfortable question of how much of this is self-inflicted.

For CSMs, this isn't just a fun infrastructure story. When your customer's team is blocked for 2–3 hours a day because their code host is down, that's a productivity conversation, a renewal risk, and a trust issue all at once.

What this means for us

The surge in code output isn't an anomaly. It's the new baseline. The platforms built for pre-AI, human-paced development are being stress-tested against a reality they weren't designed for, and some are failing that test visibly.

As CSMs, we're in a unique position. We're living this shift ourselves while also guiding customers through it. The teams that figure out how to code, when to code, and how to help their customers think about toolchain reliability are going to have an outsized impact in this next chapter.

GitHub data is sourced from: https://newsletter.pragmaticengineer.com/p/the-pulse-ai-load-breaks-github-why?utm_source=post-email-title&publication_id=458709&post_id=196807137