The Tools Nobody Built (Until Now)

There's a particular moment in engineering that never quite makes it into the tech press: when something breaks in production, the postmortem reveals a gap nobody expected, and someone realises a tool should exist to catch it next time. This morning, we're looking at three stories about that exact gap-the thing between "this problem is known" and "we actually solved it."

The Quiet Engineering Gaps

A developer on Dev.to has documented ten open-source projects that solve problems most teams have already hit: GitHub going dark mid-deploy, AI bots hijacked by prompt injection, container layers silently rotting with CVEs, Kubernetes pods starving each other. None of these are new problems. What's novel is the specificity. These aren't general observability platforms or commercial products requiring a sales conversation. They're narrow tools you can drop into a workflow and have working in an hour. The piece walks through each one-GitHub Reliability Shield for automatic failover, IssueGuard for prompt injection attacks, ContainerScan Pro for vulnerable base image layers, and seven others. What they share is a ruthless focus on solving one thing well, which is exactly what production systems need.

AI Building Things (And Getting Surprised By Its Own Work)

Meanwhile, a developer gave an AI agent a straightforward brief: build a resume formatter that actually works without paywalls or email capture. The result is OhMyDoc, a Next.js app that does exactly that-paste text left, see it formatted right, export to PDF. But here's the interesting bit: the AI built an XML editor nobody asked for, letting you tweak specific phrases without another AI round-trip. It over-engineered in exactly the right direction. The project ships, it solves the problem, it costs zero monthly to run. The takeaway isn't that AI replaces developers-it's that AI is good at velocity when the scope is tight and the problem is real.

Creative Directors Learning to Work With Infinite Tools

In Seattle, creative studios working with AI video tools are discovering something counterintuitive: AI doesn't necessarily save money. To get one usable shot, you might need 1,000 generations. But it does change what humans do. Brice Budke and Zeek Earl from Packrat and Shep describe AI as something that "pulls you towards the middle," always toward derivative work. The distinctiveness comes from human judgment, taste, and perspective. When they combined handmade production design with AI-assisted animation on the Kiro project, Zeek said he "felt in control of AI for the first time." That's the pattern: AI handles busywork, humans provide the vision. The psychological challenge is real too-infinite possibility means you never feel done. But their hope is the same as it always is: lower costs unlock weirder, more interesting human creativity.

The thread connecting these three stories is patience with the specific over excitement with the general. Tools that solve one production problem well. AI that builds to brief when the brief is tight. Creative tools that augment rather than replace judgment. None of it is significant. All of it is useful.