Notes from building AI systems, Python services and the tooling around them.
CTV is adopting outcome-optimization models faster than it is fixing the content signals those models are supposed to learn from. The result won't show up as a broken dashboard. It will show up in the price of premium content.
7 min read
What this small LangGraph starter leaves out, a concrete list of next steps, and what I learned and would do differently.
4 min read
Every failure from building a local LangGraph agent: a stale ChromeDriver that made the model say a site was down, invented dates, and more.
10 min read
Replace pretty_print with a compact streaming display and silence noisy MCP server logs with a small wrapper, without touching the servers.
8 min read
Swap LangGraph's in-memory checkpointer for SQLite so a local agent remembers across restarts, then trim old tool results so a 7B model keeps up.
9 min read
Pause a LangGraph agent before it writes files: interrupt(), Command(resume=...), a real approve and deny run, and the bugs I hit along the way.
12 min read
A 21-case eval for a small-model router: how it works, what a 100% score does and does not prove, and a sanity check that shows it can fail.
10 min read
Add a router node and one agent per tool group in LangGraph so a 7B model sees only the tools it needs. Covers conditional edges and what still goes wrong.
10 min read
Rebuild the prebuilt create_agent loop with LangGraph's StateGraph: state, reducer, nodes, edges, ToolNode, a checkpointer, and what the prebuilt hid.
8 min read
A 7B model with many tools invents dates, miscounts and misreports errors. What I cut, what the system prompt says line by line, and what stayed broken.
10 min read
Wire a 7B model in LM Studio to three MCP servers with LangChain's create_agent, and read the tool calls it makes on a real run.
9 min read
Four measurement providers reported 7.4M to 19.2M households for the same campaign. Warner Bros. Discovery's Unified Reach Paradox shows identity resolution...
7 min read
A driverless Waymo drove past a "Road Closed" sign into Denver’s South Pearl Street Farmers Market. Nobody was hurt, and nobody could be cited. That second...
10 min read
"AI agent" has become one of those phrases that means everything and nothing. Marketing decks throw it at every LLM that calls a function. Vendor pages...
5 min read
A fifth of London's jobs face significant AI exposure according to a Greater London Authority report by economist Jeff Dwan-O'Reilly. The numbers deserve...
4 min read
Two of the most consequential AI infrastructure stories of April 2026: Resonac formally opened the first US center dedicated to advanced semiconductor...
9 min read
Quick-reference guide to all major technologies and concepts on the CCA exam: Agent SDK, MCP, Claude Code, Claude API, Message Batches API, JSON Schema, and...
11 min read
Four hands-on labs mapping to CCA exam domains: build a multi-tool agent, configure Claude Code for a team, implement a structured extraction pipeline, and...
13 min read
12 worked exam questions with full distractor analysis. Each question covers a realistic production scenario with correct answer explanation and analysis of...
15 min read
Build a reliable data extraction pipeline with JSON schema design, validation-retry loops, few-shot examples for varied document formats, and...
4 min read
Integrating Claude Code into CI/CD pipelines using non-interactive mode, JSON output formatting, and the Batch API for cost-optimized asynchronous review.
4 min read
Applying Claude Code to developer workflows: iterative refinement, context management in large codebases, and team-scoped configuration patterns.
4 min read
Design a coordinator-subagent research pipeline with parallel execution, structured output, error propagation, and provenance tracking across sources.
4 min read
Using Claude Code for code generation workflows: plan mode for complex refactors, CLAUDE.md hierarchies, and slash commands for team automation.
3 min read
Build a customer support agent with get_customer, lookup_order, process_refund, and escalate_to_human tools. Covers prerequisite gates, escalation criteria,...
3 min read
Managing conversation context across long interactions, error propagation in multi-agent systems, human review workflows, and provenance tracking during synthesis.
10 min read
Designing effective tool interfaces, structured error responses, tool distribution across agents, MCP server configuration, and built-in Claude Code tools.
9 min read
JSON schema design, tool_use and tool_choice, few-shot examples, validation-retry loops, and multi-pass review architectures for the CCA exam.
9 min read
Complete coverage of CLAUDE.md hierarchies, .claude/rules/ glob patterns, slash commands, skills, plan mode, CI/CD integration, and the Batch API.
9 min read
The largest CCA exam domain. Master agentic loops, multi-agent coordination, subagent configuration, workflow enforcement, SDK hooks, task decomposition,...
12 min read
A high-level meta-reference covering all core technologies tested on the CCA exam — Agent SDK, MCP, Claude Code CLI, and key cross-domain themes.
2 min read
Your starting point for the CCA Foundations exam: domain weights, scenario overview, study strategy, and progress tracker.
3 min read
OpenAI promised to cure cancer. Denver Public Schools just blocked them from helping with homework. The gap between those two statements tells us something...
2 min read
A personal reflection on building Flask applications that are maintainable, secure, and enjoyable to work with over time.
7 min read
The AI conversation has been dominated by model capabilities for so long that we've missed what's actually happening on the ground. While the industry...
8 min read
The AI gold rush has created a peculiar hiring paradox: everyone wants "AI engineers," yet the most critical infrastructure challenges facing AI adoption...
5 min read
Have a project in mind, or a question about something here?