The Solo Stack — field notes for solo builders
Building products alone, with AI.
Matt’s hands-on notes: local models, owned infrastructure, tool reviews with the gotcha included, and what breaks.
One confirmation email at launch — nothing sends unless you confirm.
Latest articles
82 articles, newest first
Seven Local Models, Zero Clean Passes
The fastest model missed required behavior. The strongest extractor still failed a gate. A small acceptance matrix changed the model-selection question.
DGX Spark at 64 Concurrent Requests: Throughput Is Not Latency
Two vLLM recipes converged near 558 aggregate tokens per second at concurrency 64. The individual request got four times slower. Both numbers are the benchmark.
I Fine-Tuned the Model. The Baseline Still Won.
LoRA, DoRA, and a revised RAFT adapter all missed the deployment bar. The larger lesson was not about training. It was about freezing the serving path before…
Get new posts by email — first
Everything here is free to read — 82 published field notes on local models, owned infrastructure, and tool reviews. The newsletter is in the works; the waitlist gets the launch email first.
The Solo Stack is written by Matt — building products solo with AI, on his own infrastructure. If a claim isn’t backed by experience or a measurement, it doesn’t ship.
Not sending yet: joining stores your address on the waitlist. One confirmation email at launch — nothing sends unless you confirm.
Google Delayed Gemini 3.5 Pro: Contingency When Your Stack Bet Slips
When a flagship slips on coding and long-horizon work, the lesson is not gossip. It is a failover table you should have written before the delay hit the news.
Default AI Remix of Public Posts: Creator Rights for Solo Brands
When public social posts are opted into AI remix by default, solo brands need a settings checklist — not vibes. Opt-out is not always retroactive.
1000 Tokens/Second Coding Agents: Why Latency Changes the Workflow
When coding models stream near-interactive speed, the bottleneck moves from generation to verification. Solo builders who do not redesign the loop just fail…
Near-Opus Pricing and the Tokenizer Tax
Intro pricing can look like a gift until long agent traces burn more tokens per task. Solo builders should price accepted outcomes, not million-token list…
Tracking the Local and Open Frontier in 2026
Local models, open weights trained off NVIDIA rails, and funded open-model clouds changed the economics. Here is how solo builders place the next workload…
Coding Models Trained on IDE Traces: Efficiency and Contamination Honesty
Models trained on real agent-IDE interactions change coding economics. Self-disclosed benchmark contamination is a trust signal — and a reminder that private…
ChatGPT Work + Sites vs Owning the Stack
Hosted vibe-coding and chatgpt.site-style publishing are legitimate MVP paths. Ownership is a different product with different failure modes. Here is the…
GPT-Live Is Full-Duplex: Product Interfaces When Voice Stops Taking Turns
When a voice model can listen and speak at the same time, your product UX changes. So do your audit requirements. Here is when full-duplex helps a solo product…
Meta's First Paid Model API: When a Solo Builder Should Open That Account
A new paid agentic API with long context and computer-use claims is worth a trial protocol — not a migration. Here is the GO / WAIT / SKIP checklist I use…
MCP Is the Integration Layer Now
Managed agents and remote MCP turned chat polish into a commodity. The scarce skill for solo builders is tool contracts, permission boundaries, and…
OpenAI Is Buying Engineers, Not Just Models
When a frontier lab spends billions on forward-deployed engineers, it is not a model announcement. It is a market price for implementation. Solo builders…
The Model Ladder After GPT-5.6: Sol, Terra, Luna as Economics, Not Branding
OpenAI shipping Sol, Terra, and Luna is not a personality pack. It is a cost-control API. Here is how solo builders should re-route work after the three-tier…
Get new posts by email — first
Everything here is free to read — 82 published field notes on local models, owned infrastructure, and tool reviews. The newsletter is in the works; the waitlist gets the launch email first.
The Solo Stack is written by Matt — building products solo with AI, on his own infrastructure. If a claim isn’t backed by experience or a measurement, it doesn’t ship.
Not sending yet: joining stores your address on the waitlist. One confirmation email at launch — nothing sends unless you confirm.