Here's the paradox nobody warned us about: AI made it cheap to generate ideas, draft plans, and start projects. It did not make it cheap to track them. The result is a generation of developers sitting on 5-10 half-built things with no coherent way to explain status—to stakeholders, to teammates, or even to future-themselves.

The Bottleneck Moved

For most of software engineering history, writing the code was the hard part. Planning was easy: think it through, scope it, spend two weeks building it. AI inverted this completely. As author and developer yasiru writes on DEV.to, they can now scope a feature with Claude in 20 minutes and have a working prototype by afternoon. That's genuinely transformative—but the bottleneck didn't disappear. It moved to knowing what you're actually working on. The symptom manifests concretely: scattered FEATURE_X_PLAN.md files across repos, some projects started enthusiastically and quietly abandoned around the 60% mark, ideas AI-planned that never moved past the planning doc, dependencies between half-built things that were understood when written and completely forgotten two weeks later. "At one point I had 38 plan files at the root of one repo," yasiru notes, "none of which I could quickly classify as 'alive' or 'dead' without re-reading them."

Why Traditional Tools Don't Fit

The obvious response is: use project management software. Jira? Tried it—lasted a week. Linear? Same outcome, faster. Notion? Lasted longer because it felt closer to writing, then sat unused after three weeks anyway. The pattern yasiru identifies cuts deeper than UX complaints: these tools are built for teams shipping a small number of larger things in coordinated fashion. That's not what AI-assisted solo work looks like. "AI-assisted work generates a larger number of smaller, parallel, shifting things with unclear states and frequent context switches," yasiru writes. "Different shape of work needs a different shape of tracking."

The Fix: Plans

Instead of forcing square pegs into round holes, yasiru built their own convention called Plans—three directories (active/, shipped/, superseded/), one file per feature with seven fields of frontmatter (status, priority, owner, type, depends_on, blocks, last_updated), and a STATUS.md that answers "where are we right now" in one place. The key insight is the depends_on and blocks fields: plans aren't independent, they have a dependency graph, and making that explicit lets you see "I can't start B until A ships." The dashboard (roadmap.html) reads auto-generated JSON from these markdown files and renders an interactive Gantt view plus dependency graph plus filterable cards—all static HTML, no server, no account, no login. This is the thing yasiru sends to stakeholders. "They open a URL, they see what's happening," yasiru explains. "No tool training, no permissions, no monthly seat cost."

The AI Bonus Nobody Expected

The format wasn't designed for AI consumption—but because it's plain markdown with predictable structure, AI coding assistants read it natively without any integration, MCP server, or plugin. Claude Code, Cursor, Antigravity, Windsurf all open the files and immediately know what's in flight. "When I start a Claude session now, it reads STATUS.md and the relevant plan files," yasiru notes. "It already knows the state. We start working in two messages instead of fifteen."

The Real Insight

"AI didn't fix my planning problem. AI made my planning problem urgent. The fix was structure I should have had all along." That's the thesis, and it cuts to something real about how development workflows are shifting. As spec-driven development becomes more prevalent—GitHub's Spec Kit released late 2025 handles per-feature specs well—the portfolio layer becomes critical. "Spec Kit answers 'how do I build this one feature well.' Plans answers 'what are the 8 features I'm juggling, which are alive, which are dead, which are blocked, and what's next.'"

Key Takeaways

  • AI coding assistants have inverted the development bottleneck: building is cheap, tracking remains expensive
  • Traditional PM tools (Jira, Linear, Notion) were designed for team coordination on fewer, larger deliverables—not solo devs with many parallel shifting projects
  • The solution lives in-repo as plain text, eliminating friction: no browser tabs, no separate apps, just markdown in the editor
  • AI assistants naturally read structured markdown conventions without any integration work

The Bottom Line

This is a real problem and yasiru built something genuinely useful. If you're running Claude Code or Cursor and wondering why your "what are you working on" answer requires an hour of archaeological research, the 300 lines of code that make up Plans might be exactly the structure you've been missing. MIT licensed, one curl to install, lives in your repo.