Title: Blog — field notes on AI-native engineering | Glideship
Source: https://glideship.ai/blog/
Summary: Essays on building an AI-native engineering org: Agent-Driven Development, why seniors still own the hard 20%, and hiring for the judgment AI can’t generate.

Field notes

# Building an AI-native engineering org, in the open

Essays on what changes when agents write most of the code — the methodology, the judgment that doesn't automate, and the system I open-sourced to hold it all together. First published on dev.to; collected here.

[Building Glideship We Loved Linear. We Just Didn't Have a Tracking Problem. Linear is the best issue tracker I've used — so this isn't a takedown. It's what happened when AI started writing most of our code and we realised a tracker, however beautiful, still assumes a human writes every line. We didn't need a faster tracker. We needed a source of truth the agents could build against. 14 Jul 20266 min read AIToolsProject ManagementEngineering Read the piece](https://glideship.ai/blog/we-loved-linear-didnt-have-a-tracking-problem/)

## More writing

[On method Write the Spec First. Everything Else Gets Easier. I used to think specs were busywork. Then AI agents started writing my code, and the spec became the whole job. A short case for deciding what "done" means before anyone — human or machine — starts building. 4 Jul 20264 min read Spec-Driven DevelopmentAIEngineeringMethodology Read the piece](https://glideship.ai/blog/write-the-spec-first/)

[Building Glideship I Replaced Scrum, Jira, and Our Wiki With 12 AI Agents on a Mac Mini More than half the code we ship is AI-generated — so I stopped bending fifteen-year-old Agile process around an AI-native team and built my own. Twelve agents, one living document per feature, and the whole org rendered as a 3D grove. It runs on a Mac mini. 8 Jun 202613 min read AIOpen SourceDevOpsMethodology Read the piece](https://glideship.ai/blog/replaced-scrum-jira-wiki-with-12-ai-agents/)

[The 80/20 series · Part 2 How We Hire for the 20% AI Can't Do If AI writes the clean 80%, an interview that tests clean-code-under-pressure is testing the wrong thing. So we stopped asking candidates to write code — we hand them AI-written code that looks fine and quietly costs money three weeks later, and watch what they catch. 2 Jun 202610 min read HiringAICareerEngineering Read the piece](https://glideship.ai/blog/hiring-for-the-20-percent-ai-cant-do/)

[The 80/20 series · Part 1 AI Agents Are Great at 80% of Our Code. The Other 20% Is Why We Still Need Seniors. We let AI agents loose on an FCA-authorised payment platform. They crushed the boring 80%. Then they silently broke the 20% that moves money — the illegal state transitions and idempotency a senior encodes in their bones, not their code. 28 May 20266 min read AIFintechWeb DevStartup Read the piece](https://glideship.ai/blog/ai-agents-great-at-80-percent-seniors-for-20/)

## The methodology behind the writing

[Explore the platform](https://github.com/mickyarun/glideship)

Every essay here traces back to one open-source system. See how Agent-Driven Development actually works. `npx glideship init`
