
My Agents Are Building a Game World
Testing how generated assets come together in a game world.

Testing how generated assets come together in a game world.

Building a dashboard that shows what still needs attention.

Testing AI researchers with distinct personalities and simulated investments.

Comparing three local models for our research agents.

Organizing generated media so we can find and review it.

Testing real software with simulated customers before connecting live services.

Spending less time steering agents back to the plan.

Teaching our StarCraft II bots to beat the Elite AI.

Turning hiking photographs into playful videos with local AI.
A working playbook for short-form AI video in 2026. Which models for which shots, image-to-video over text-to-video, last-frame chaining, native synced audio, and the multi-shot generation that just changed the workflow.
A field guide to Agent SDK flows and runbooks for builders. What each one is, when to use each, why Agent SDK flows are brittle on dynamic work, and how to have Claude Code write a runbook for you.
The economics of building changed, and the projects you killed two years ago got quietly resurrected. Here's the framework for deciding what to pull off the shelf first.
Running MoE models 24/7 on DGX Spark will thermal shutdown your GPU. One nvidia-smi command fixes it with zero throughput loss.

A question from my wife about ChatGPT reminded me that most people still think AI is one thing. If we want to help them, we have to be teachers too.
We used AI to create an entire fictional country music artist from scratch. Original songs, a consistent voice, branded website, the works. Here's what that taught us about creative AI.
The role changed faster than the talent pool. What companies need isn't an ML researcher. It's a technical architect who's been building AI-first for years.
AI-assisted development collapsed costs by an order of magnitude. Most companies are still budgeting like it's 2023 and missing the window.
A solo founder built a platform, hit $1M ARR in three weeks, and sold to Wix for $80M cash. He never raised a dollar. This is what happens when building becomes free.
Companies buy AI tools, keep old processes, get 15% gains instead of 10x, and blame the tools. The org structure is the bottleneck.
Being a great traditional engineer doesn't make you good at AI dev. The gap between the best and worst AI developers is the widest skill gap in the history of software.
Vibe coding went from Karpathy tweet to Stanford syllabus in thirteen months. When mortgage bankers are learning it, the tipping point is ancient history.
84% of Uber's developers adopted agentic AI tools in one quarter. No committee approved it. No pilot program blessed it. The enterprise adoption debate just ended.
Stop hand-tuning agent personalities. Define the trait space, run variants, breed the winners. Intuition doesn't scale.
Multi-agent systems drift toward monoculture through shared memory. Here's how we caught it and what we did about it.
Most agent systems fail because there's no structure around what agents do, when they do it, and whether it was any good.
Long-running agents degrade as context fills up. Shorter contexts, structured memory, and tiered consolidation keep them sharp.
We replaced a 200K-star framework with 5,000 lines of Python. Agentic coding killed the one argument against forking.
How we unified a laptop and desktop workstation into a single dev environment with GPU offloading, workspace sync, and a gaming mode toggle.
Nine agents ran around the clock this week. Most produced garbage on Monday. By Friday, after four debug cycles, the architecture looked completely different.
We built an agent pool architecture with 4 workers and an ops agent running 24/7 on systemd. Swappable personality profiles, isolated state, pennies per day. Like a staffing agency for AI.
Software stocks are in freefall. The panic is indiscriminate, but the underlying disruption follows a clear pattern—and creates massive opportunity for builders.
The journey of deploying MiniMax M2.1 across two NVIDIA GX10 units. What broke, what worked, and what others can learn from it.
Prompt injection drops from 50%+ success rate to under 2% with one structural change. Here's the pattern and how to implement it.
Building a local AI assistant to help run my business. The tech is ready and prices are going up.
Setting up Clawdbot on an RTX 5090 laptop instead of a Mac Mini, with plans to let it argue for its own hardware upgrade.
The competitive advantage isn't working harder. It's not needing willpower to do the work at all.
AI is the new Manhattan Project. Every breakthrough needs physical infrastructure. Silver sits at the center of batteries, solar, military tech, and data centers—and we've run deficits for five straight years.
Software devs stopped writing code and started directing agents. The same shift is coming to battlefield robotics—and it's coming there first, before commercial applications.
One person can do what 50 did pre-AI. Stop touching code and let the agent do it.
If 1 engineer can do what 50 did, what happens to everyone else? The bottleneck was never ideas—it was cost.
From brittle scripts to a self-correcting pipeline. How we turned video generation into a multi-phase agent that researches, writes, critiques, and improves its own work.
My workstation was 93% full with a forgotten 4TB NVMe sitting empty. Claude Code found it, set it up, and cleaned house while I watched.
The architecture doc is the product. Code is just the implementation. Here's how we actually work with Claude Code.
Our social media bot sounded like garbage. We couldn't explain why. Here's the agentic debugging process that's fixing it.
How we gave our AI real-time context with Brave Search and URL fetching. The integration details nobody talks about.
The pattern: cheap models filter, expensive models generate. A simple architecture change that dramatically reduces API costs.
What happens when building products becomes more fun than gaming. A month of shipping.
The workflow that lets us ship production software in days. The secret: stop typing code yourself.
How we built an MCP server so Claude Code can query our internal knowledge base instead of re-explaining context every session.
An AI assistant that runs in your infrastructure, keeps your data private, and costs a flat monthly fee.
Try a different word, or clear your search to see all posts.