You Can't Vibe-Code Reliability
jalotra@jalotra
The idea
Code is getting cheap to write, but running it reliably at scale is not, especially as more AI-generated code ships half-reviewed. I want exposure to whoever has to keep that code running and clean up after it, so this leans into monitoring and observability, real database infrastructure, battle-tested platforms, tools for shipping code safely, and a smaller slice of security. It is a fairly concentrated basket across many names rather than single bets, with some cash held back, built for a one to two year view on where the breakage and the bills show up first.
What it owns
- As more AI-written code ships to production, companies need to watch it harder for breaks, and these platforms are the default tools for spotting and fixing outages fast. · DatadogIt watches production software for breaks and slowdowns, the default tool for companies serving customers with AI-written code., DynatraceIt watches the huge, messy systems at big banks, insurers and government agencies, where an outage costs real money and AI coding cannot fix it fast., ElasticWhen something breaks in production, the first move is searching the logs, and it stores them for a lot of companies to dig through fast., PagerDutyIt becomes embedded in incident response workflows, making retention dependent on whether outages and complexity continue to drive adoption across teams.23%
- Big companies trust systems with decades of fixes built in, which AI coding over a weekend cannot copy, so these platforms stay essential as code velocity rises. · MicrosoftIt sells both the AI coding tools through GitHub and the Azure cloud with decades of hard lessons built in, so it wins whether vibe coding takes off or stalls., ServiceNowIt logs IT problems, outages and fixes at big companies, the paperwork side of keeping systems up for a lot of customers with more code going live., OracleBanks, airlines and hospitals run on its databases precisely because they cannot afford to break, which is the take's point in its purest form., SAPNobody vibe-codes the payroll, ledger and supply chain systems for a global company, and it runs those systems for a huge share of them.22%
- Code nobody fully reviewed leaves security holes, and these platforms close them by protecting apps, logins, and network access without slowing down deployment. · CrowdStrikeIt protects the laptops and servers where attackers get in through code nobody really reviewed, which becomes a bigger risk as AI writes more software., CloudflareIt sits in front of apps to block attacks and soak up traffic spikes, a cheap way for a company to protect software it didn't fully review., ZscalerIt controls who and what inside a company can reach which systems, which matters more when there's a pile of quickly built internal apps nobody vetted., OktaLogins and permissions are easy to get wrong in a vibe-coded app, and it sells that piece done right for companies adding more apps and tools.18%
- Easy-to-start apps become hard to run once they serve millions of users reliably, and these databases handle the work that can't be vibe-coded away. · MongoDBIts Atlas database holds up under real load once a vibe-coded app gains a lot of customers, a common choice for developers building at scale., SnowflakeIt bills by consumption of its data platform, so usage patterns in production environments directly determine revenue.15%
- When code is cheap to write, the bottleneck shifts to testing, reviewing, and releasing it without breaking things, and these tools run that pipeline. · AtlassianIts Jira and service desk track work and incidents, where engineering teams manage the mess as more AI-written code goes live and breaks things., GitLabIt runs the pipeline of reviews, automated tests, and security scans before code ships, which matters more as AI writes more of the code., JFrogIt stores and vets every piece of software a company builds before it goes live, so more code being shipped means more to scan for bad packages.12%
- Cash10%