Saturday in the Life of the AI Enhanced
What follows is a few hours of work by just me, not a team, for a half day on Saturday. Consider it a companion piece to the Dorsey 40% reduction in force seen from the other side - the expansion in ability.
I set up agentic workflows… and went for a walk.
Later, coding continued while I played with my grandson and watched TV.
Multiple work streams ran simultaneously.
Betsy — my OpenClaw autonomous bot — is back from the dead (I killed her during a security hardening exercise… long story).
Here’s what got done.
⸻
1. Snowflake Client Work — Production Engineering
For one client engagement:
• Had Codex review the CSV → Snowflake automation workflow I built.
• Had Codex design a JSON ingestion strategy, including nested structures.
• Since client files weren’t accessible, Codex generated realistic .ndjson sample data, zipped as .gz.
• Built and executed a pipeline to load JSON, NDJSON, and GZ files into Snowflake.
• Had Codex produce client-ready documentation explaining how to operate the pipeline going forward.
That alone is multiple days of engineering + documentation.
⸻
2. Proposal: Waterfall → Agile Shift
I’m on a project drowning in discovery and human coordination complexity. Planning fixed dates right now is fiction.
So:
• Strategized the positioning and argument structure with ChatGPT.
• Clarified why backlog + burndown is the correct operating model.
• Framed the case: we already know “the date” won’t happen — pretending otherwise helps no one.
• Handed the structured thinking to Claude CoWork to draft the formal proposal.
Strategic narrative + operational shift — done in hours. I assure you, I could NOT create as nice of a Word document complete with diagrams. Certainly not as quickly.
⸻
3. Local Text-to-Voice (Kokoro TTS)
I selected Kokoro TTS as my local voice engine.
• ChatGPT generated an implementation plan.
• I asked Betsy if she could execute it autonomously.
• She spun up a sub-agent to implement it — while she and I worked on other things.
Parallel execution. No babysitting. The bot enhancing itself. Not 100% completed yet.
⸻
4. LLM Failover Problem → Product Opportunity
My ChatGPT subscription hit usage limits during recovery work. Auto-failover to Kimi K2.5 didn’t trigger.
Had Betsy write up a spec to do this with a python app from Linux. Had Kimi Code code while I walked the dog. Had Codex review and improve while I ate lunch.
⸻
5. Cross-Machine AgentFlow Deployment
Betsy runs on Linux. My init-agent framework lives on Mac.
So:
• Had Codex build a Linux executable to deploy AgentFlow on any Linux project.
• Used it to take the above project and enable AgentFlow methodology
Standardizing my agent orchestration layer across environments.
⸻
6. Meanwhile… Betsy Is Running Hourly Autonomy Workstreams
While I’m doing all of the above, Betsy is executing her own structured hourly loops:
• Storytime Sleep Fiction (Episode 2 production pipeline)
• Autonomy Backlog Governance Refinement
• BAAi (Business AI evaluator artifacts + SLA continuity structures)
• Partnership Improvement (Betsy × Lee operator pattern logging)
• Daily AI Intelligence Briefing system
• Post-security-hardening recovery and stabilization
She:
• Advanced production artifacts
• Refined governance cadence systems
• Built comparator and SLA action cards
• Expanded operator learning indexes
• Hardened operational reliability
All in background loops.
⸻
This wasn’t “vibe coding.” This was orchestration.
Strategic thinking.
Delegation.
Parallelized execution.
Multiple LLMs in role.
Agents generating artifacts while I lived my life.
A few hours. One human. A stupendous amount of output.
AI isn’t replacing engineers with AI. It’s turning one experienced engineer into a small, coordinated firm.
