Capital de Tokens
Ideas, analysis, and independent articles from the channel and its episodes.
Blog RSSChat Jimmy demonstrates a radical idea for AI inference: specializing silicon around a model. This is how Taalas HC1 works, why it reaches extreme speeds, and what it could mean for agents and AI systems.
What reliability engineering is, how it relates to SRE, and why availability, observability, resilience, retries, checkpoints, and idempotency are essential in modern systems and AI pipelines.
AI is changing the CTO's job: from scaling teams and managing code production to designing engineering systems where humans and agents collaborate under continuous evaluation, governance, and verification.
How Strategy, Adapter, Dependency Injection, Circuit Breaker, Factory, and Facade combine with timeout, retries, backoff, jitter, fallback, and metrics to build more robust AI clients.
OpenAI adds Computer History to ChatGPT for macOS. What it records, what it does not, why it changes the relationship between the assistant and the desktop, and how it fits into the evolution toward persistent agents capable of continuing real work.
Google moves A2A under the specialized umbrella of the Agentic AI Foundation. What changes, how it fits with MCP, and why interoperability between agents could become foundational infrastructure for agentic AI.
OpenAI is turning Codex into something broader than a programming assistant: a reusable harness for building agents inside dashboards, IDEs, internal systems, and specialized products.
More marketing, more features, and more capital can make a startup worse when the diagnosis is wrong. How to distinguish strategy problems from execution problems before scaling.
Tokens, infrastructure, agents, retries, and governance: a guide to understanding what it really costs to operate AI at scale and why cost per successful task is often the better metric.
The rise of Replit and coding agents does not eliminate software engineering: it shifts human work from writing lines of code toward specifying, verifying, and directing systems that program.
The cloud is not infinite: distributed storage, erasure coding, hot/cold tiers, compression, caches, and economies of scale explain how large platforms preserve enormous volumes of data for years.
Two ideas turn an AI agent into infrastructure: decoupling it from the computer where it executes work and capturing its trajectories as data for evaluation and training.
Capital de Tokens
