Our family used Slack for years, but we recently moved to a self‑hosted chat setup (we tried Nextcloud Talk and will likely settle on Mattermost). The reasons: privacy for both conversations and attachments, and avoiding Slack’s message‑history limits and other features locked behind paywalls.
When I exported our Slack workspace, I discovered something surprising: the export didn’t include the actual files. You would think “export workspace” means exporting everything. But no—just the metadata about files, not the files themselves. There’s a note about this in the docs, but it’s easy to miss—even for someone like me who actually reads manuals.
There wasn’t an easy (or free—there are paid workspace backup tools) way to download all those files. So I decided to automate it. And rather than do it alone, I co‑developed it with Claude Code (and OpenCode—I’m still deciding which to use as my default).
The result? slack-file-exporter, built in a few focused hours over two days, then polished so I’d feel comfortable sharing it publicly—and not dread maintaining it.
What it does: Slack File Exporter downloads files only (not messages) from your Slack workspace via OAuth and organizes them by channel/DM with readable names. You authorize users in a tiny web UI, then download via the UI or CLI; it handles rate limiting, logging, and basic verification. It requires a Slack app with user scopes and an HTTPS redirect (cloudflared works); see the README for setup and commands.
Read more: https://github.com/ryangaraygay/slack-file-exporter#readme
Why Build Something Shareable?
Sure, this started as a one‑off need. I could have hacked together a quick script just for myself. But working with an AI coding assistant changed my approach. I thought: why not make something with reasonable quality that others might actually use?
That’s the subtle shift AI assistants enable. I wouldn’t have bothered with modularity, proper error handling, documentation, and the other things that make code shareable. Without Claude, I probably wouldn’t have produced something I felt was worth sharing.
The Economics of AI Coding Assistants
Here’s what strikes me: the value‑to‑cost ratio of AI coding assistants is incredibly high right now—almost suspiciously high. In the right hands, a $20 Claude Pro subscription (or higher tiers) can unlock a surprising amount of leverage. And it’s not just Anthropic; GitHub Copilot, Cursor, Windsurf, Warp, Gemini—take your pick. Even one‑shot “builder” tools like Lovable can get you far. If you’re reasonably technical, one of these assistants can help you ship solutions you might never have attempted on your own—certainly not for $20/month.
I suspect many providers (the model vendors themselves, not necessarily tool makers) are subsidizing usage to drive adoption. That creates a window of opportunity for us as developers and builders.
This is where fixed‑price subscriptions shine. Even better? Leverage free tiers and open‑source models. In many domains, open models (Qwen, GLM, MiniMax, Kimi, and more) are catching up fast—and sometimes surpassing closed models—for little to no cost.
The Window Might Not Last Forever
Some providers have already pulled back unsustainable offerings or tightened limits. But with new high‑quality models announced constantly, competition is keeping costs down—for now.
I don’t know if competition keeps prices low long‑term, or if the reality of compute and energy costs will push prices up and consolidate the market.
What I do know: while it lasts, use this window to solve real problems—no matter how small.
Yes, we should try new tools—it feels like a new IDE, CLI, or model drops daily. But don’t forget to actually solve real problems. That’s the whole point.
Don’t Just Flex—Build Real Solutions
AI tools are fun—sure. But the real value is solving problems you might otherwise avoid, simple or complex. I set up my Proxmox VE homelab slowly and carefully with Ansible—then Claude/OpenCode accelerated what I could do with it. I migrated my WordPress blog to Hugo with static hosting on Cloudflare Pages in under an hour. I’d been putting that off for years. Those are posts for another day, but I haven’t been this excited about technology in quite a while.
I didn’t build slack‑file‑exporter to show off (though it was fun), or to pretend I wrote every line alone, or to claim I “know” AI. I built it because I had a real problem. AI made it feasible to solve it properly—not just with a quick hack, but with something modular, documented, and shareable.
My Challenge to You
Take advantage of co-developing with AI assistants while costs are still low. Use these tools to:
- Solve problems you’ve been putting off
- Build that side project you’ve been thinking about
- Automate that annoying task at work
- Create something useful for others, not just yourself
Don’t just use AI to learn or experiment (though that’s valuable). Use it to ship real solutions. The barriers to building quality software have never been lower.
Thanks, Claude
Yes, I could have written a quick one‑off script to call APIs. But I wouldn’t have bothered with modularity, quality, and sharing—if not for Claude.
Thanks, Claude. And thanks to everyone pushing these tools forward.
Now go build something while it’s still this cheap.
Check out the project: slack-file-exporter on GitHub