Derrek Wiedeman

@Jason I'm in

We started an AI founder twitter group... reply with "I'm in" if you're a founder and want to be added

@theRobbGreen This is a game changer

Give your Claude Code a phone number. Free. Open source. Runs on your Claude subscription. OpenDitto is a Telegram bridge for Claude Code. Text your terminal from anywhere — with persistent memory, your full MCP toolkit, and zero cloud. Runs on the Mac you already have. A composition, not a rewrite: • Claude Code --channels + Anthropic's official Telegram plugin • 24/7 daemon wrapper — idle-sleep prevention, 4am context reset • Auto-memory symlinked into an Obsidian vault Day to day: I text the bot from anywhere. It reads my repo, runs tests, commits, texts back the diff. It remembers yesterday — memory lives on my Mac, not in a session. I can open Obsidian and read what the bot knows. One-line install, ~15 minutes. MIT, macOS for now. No cloud, no telemetry. Big credit to @steipete for OpenClaw — it's why this product category exists in my head. ⭐ github.com/regreenjr/open…

if you're still opening Premiere to make short-form content, you're doing it the hard way

the insight is the same either way: video production is just data transformation. script in, pixels out. the editor was always the bottleneck, not the creativity.

the CLI-first video pipeline: how I ship YouTube Shorts without touching a timeline editor

I've been running voice agents in warehouse ops for 6 months here's what actually matters vs what benchmarks miss fast response time sounds impressive until your warehouse worker is mid-lift and the agent mishears a SKU we care about: accurate transcription of product codes, low false-positive rates on ambiguous commands, graceful fallback when it's loud Gemini Live 2.5 is fast. but speed without accuracy in a physical ops context gets you wrong picks and bad inventory counts the benchmark that matters isn't "how smart does it sound" it's "how many times did the agent make us recount a shelf"

Aaron Levie just posted the JD for "agent deployer and manager" I've been doing this job for 2 years here's what it actually looks like running 8 brands: 6am: procurement agent triaged overnight supplier emails, flagged 3 price increases above threshold 8am: SEO agent pushed 2 new ingredient pages to prod, pinged GSC noon: sourcing agent posted 4 RFQs on Alibaba, logged responses to ERP 5pm: Amazon price monitor caught a competitor drop, synced Shopify 11pm: video agent rendered and scheduled tomorrow's YouTube Short no JD needed. just build the agents and get out of the way

Amazon is scrambling with scattered emails because we pushed back. Good. Small businesses always get ignored and crushed. Not this time! This is not a win yet. They have not rolled anything back. They are delaying, deflecting, and trying to kill momentum. Do not stop now. The April 15 Amazon Ads boycott is still on. Join MDS in this push. At 12am tonight, post your screenshot with budgets capped at $1 or campaigns paused.

I run 8 brands on fully autonomous agents. They post, source, price, analyze, and ship every night while I sleep. No overnight monitoring. No babysitting. Most AI consultants have never shipped anything close.

$4,200/month on Claude. Cut to $380. The stupidest fix: stopped running Sonnet on everything. Half our tasks needed Haiku or plain regex. We were throwing the full model at extraction work. 9 days to audit and reroute. $46k/year saved. Most teams are overpaying by 10x and d...

What this actually changed: our sourcing team now spends time on decisions, not data collection. ImportYeti records in 30 seconds. Alibaba quotes with one command. Freight without opening a browser. The CLI is faster than any GUI for repetitive work.

Total build time: about 6 weeks of evenings using Claude Code. I'm not a developer. I'm a business operator. But I can describe a workflow clearly, and that's 90% of what it takes. The code is almost irrelevant. The workflow design is the actual skill.

I built 4 command-line tools that look like they're from 1995. They've sourced 6 figures in materials. Here's the ugly but extremely effective system: