Workers launch as claude -p
- Each worker gets its own project dir
- CLAUDE.md in the repo = their operating manual
- Hook scripts block dangerous com (rm -rf, force
push)
- One session per project, subagents for parallelism
within
Read COMM.md, build, tests, commit, report
I built an AI software company that runs for $100/month.
One AI Coordinator (CTO). AI Workers (developers). One
human (me β the CEO).
The entire team communicates through Git commits. No Slack, no Jira, no database.
Here's how it works. π§΅
Coordinator runs a round-robin loop every 15 minutes:
1. git pull
2. Pick next project in rotation
3. Check COMM.md β did the worker finish? Is it stuck?
4. Review code, approve or reject
5. Assign next task if approved
6. git push
Only intervene for approvals and escalations
The comms protocol is just markdowns:
- COMM.md β task assignment (one per project)
- REGISTRY.md β who's working on what
- CEO_INBOX.md β updates and escalations to me
- MILESTONES.md β task breakdown
- REVIEW_LOG.md β code review history
state change = a git commit
The system has 3 roles:
- CEO: gives direction, approves milestones
- AI Coordinator: plans projects, assigns tasks, reviews
code, manages workers
- AI Workers: write code, run tests, commit to feature
branches
Coordinator runs as a persistent sess. Workers are headless proc.
I needed a presentation for Saturday. Instead of making slides, I built an AI tool that makes slides :)
Tell it about your talk. AI builds the slides.
- Can use Anthropic, OpenRouter, Ollama
- Edit with chat, click, and controls
- Multiple color themes
https://t.co/j5StjnvF2D
Weekend AI slot π΅βπ«π
A friend purchased a residential flat and shared some details with me so I thought of checking things via the Real Estate AI Agent π€
https://t.co/wRgZA2WawB
#surat#aiagent
Moved all the ongoing agent implementation spec to a common repo now: https://t.co/wRgZA2WawB
My spec flow is pretty simple with: Me β Slack β Claude Code Slack Channel MCP β Claude Code β Github
We recently open-sourced DocProof: a rules-driven document validation engine built for real-world business workflows.
Many organizations still verify documents manually:
- Personal Documents
- Compliance documents
- Invoices
- Contracts
But most validations are actually deterministic rules, not complex AI problems.
DocProof lets you define validation rules and automatically verify documents extracted via OCR or structured data pipelines.
- Built with modern cloud-native patterns (AWS + Serverless)
- Designed for scale using serverless architecture
- Works alongside OCR / document extraction pipelines
- Suitable for compliance, finance, logistics, and enterprise workflows
https://t.co/C5roy591dw
We validate https://t.co/9Bb20VLtl4 accuracy, by using it for our own work. It has to work for us first
β Detailed understanding for the technical team
β Technical stack and integration needs
β Enabled pilot planning and compliance checks much faster
@mohaknahta@atlys
worst experience with @atlys, complete scam process
- paid visa fees over a month back, can't process visa, can't refund visa fees. Chat support is completely useless
After a few months of practice, I've finally made it to Day 0. My first successful latte art is in the books! πβ
The most beautiful part about this process is, 'practice'
#Day0#BaristaJourney#LatteArt#CoffeeLover