SpecBuddy

SpecBuddy is live. Review what your AI agent is going to do — before it writes a line of code. Spec → plan → step-by-step execution, all reviewable, inside your JetBrains IDE. Runs on Claude Code or Codex. Free public beta. Demo 👇 youtu.be/sEaU1ltO574

@simas_ch But doesn't the workflow itself need to be loaded then?

Do you still use a CLAUDE.md or AGENTS.md? As a Java developer with a Maven or Gradle build file, specs (I use the AI Unified Process), and architectural guidelines, I don't see any reason to have something that is always loaded in the context. What about you? unifiedprocess.ai

That doesn't give you a chance to notice your agent went completely wrong until you've burned through your whole weekly budget. Convenient for them too — nudges you toward upgrading. At least with the 5-hour cap, you get a natural checkpoint to see what's actually been done 🙂

@dark_coderz Replace those who didn't become faster

AI will make software engineers → 10x faster → Easier to replace?

When you make a mistake in a low-level language, your program crashes. When you make a mistake in a high-level language, it throws an exception and keeps running. What happens when you make a mistake in system design?

Starting a new backend service today — what stack are you picking? And if an agent's doing most of the implementation anyway, does @QuarkusIO vs @springboot still matter?

@13_narcissus As for me, the current layer is specs. I'd love to know what's next — if I found out, I'd drop everything and start working on it 🙂

@spec_buddy What will be the next layer of abstraction?

A junior dev asks you: "Should I still learn to code in 2026?" You get one sentence. Go.

@LukasCantCode Much less than a year ago. Back then, agents were far more likely to just start building some feature I hadn't even asked for.

how many of you end a prompt with “don’t write any code, i’m just asking a question” when talking to claude or codex?

@Franc0Fernand0 Like healing people with ChatGPT without being a doctor

I don't understand how you can get good at building with AI without learning to code first. If you've never built things by hand, you won't know: •What "good" actually looks like •When the AI is wrong or hallucinating •What to ask for next You're not directing the AI. You're just hoping it works.

@IsaacKing314 Always have been. Good developers try to automate as much of their job as they can — the backlog always grows faster than you can ship.

I don't fully understand why programmers seem to be the ~only group of people not angry that their jobs are being automated.

@TTrimoreau toxicity!

X is full of builders. LinkedIn is full of job titles. Reddit is full of …

@dedene Working on the next feature/refactor already. Spec-driven development is what lets me stay ahead like that.

While Cursor or Codex is working on your code, what are you doing instead?

@solopribuilds The real engineers actually prefer not typing code, but building systems. We can leave the boring part to an agent and focus on the real things.

Are you willing to go back to coding without AI?

@spec_buddy combines both: each spec is tied to its own worktree. Run multiple tasks at once, confident that the agents won't collide — and neither will you. Spec + worktree = parallel work that actually stays parallel.

Problem #2 is where spec-driven development shines. Anchor your work to a document, not a chat window. When your unit of work is a spec, you don't lose track of which task you're feeding the next prompt into. The doc is the source of truth.

🧵 1/6 How often do you run multiple coding agents on the same project at once? If you do, you've hit the wall: two agents, one codebase, endless collisions. There's a cleaner way to work in parallel. 👇

@callebtc The plan is actually the most important part, for me. Almost no tool helps you ship until you can precisely type out what you actually want to build — that's basically the whole reason I started building something around specs instead of chat.

I’ve never used any of this for coding: • MCPs • subagents • agent coordinators • memory systems • loops or graphs Agentic coding is becoming like photography: people obsess over the latest gear, optimize every inch of their workflow, and still never produce a beautiful photo worth looking at. Don't waste time on productivity larping. Plan, code, win.