Insights
Notes on building better agents
Production notes on AI agent builders, agentic workflows, LLM evaluation, and cost control.
- LLM Evaluation
Don't start with a huge eval set — turn failures into regression tests
A useful agent eval set is not finished in a meeting. Start small, capture real failures, and make sure the same mistake cannot quietly return in the next release.
7 min read - Reliability
Available isn't usable — what trustworthy agent execution takes
Adding the newest model isn't adding a line to a list — it's confirming the call actually returns. What we learned wiring up Gemini 3, and what a platform owes you for a Real run you can trust.
6 min read - Pricing
Pricing & limits — what's free and what's metered
During beta, London System Agent gives you the core loop — designing and validating agents — for free. Here's what's free, where cost starts, and how to keep it under control, on one page.
4 min read - AI Agent Builder
Hands-on: from one sentence to an API — a London System Agent walkthrough
A step-by-step walkthrough: compile your first agent graph from a single sentence, validate it for free in Stub, then run real-mode evaluation and call the same agent over an API.
9 min read - AI Agent Platform
Why everyone is building AI agent platforms: the next fight is not just models
Why OpenAI, Google, Microsoft, Salesforce, LangChain, and CrewAI are converging on agent platforms — and why that creates a new learning curve for teams.
8 min read - AI Agent Builder
AI Agent Builder for Production — introducing London System Agent
The hard part is no longer generating an agent demo. It is knowing whether the agent is ready to run. London System Agent connects graph compilation, evaluation, cost control, and API execution.
5 min read - Agentic AI Workflow
Agentic AI workflow design: why graph-based agents beat long prompts
Why agentic AI workflows are easier to debug, meter, and evaluate when represented as nodes, edges, and state instead of one long prompt.
6 min read - AI Agent Cost Control
AI agent cost control: budget caps and credit gates without runaway spend
How London System Agent manages LLM spend when every build calls a live model: per-run budget caps, eval budgets, a prepaid-credit gate, and a per-step cost panel.
5 min read - LLM Evaluation
LLM evaluation for AI agents: regression tests before deployment
An agent finishing is not the same as an agent being correct. Here's how structure checks, golden datasets, and policy gates catch regressions before production.
11 min read