We built an internal assistant that researches crypto markets and drafts a briefing for analysts, using a planner and specialised sub-agents. The lessons: isolating sub-agent context, a file scratchpad instead of one huge conversation, cost and tool caps, and evals. It does research, never trading.
Agent skills, plugins and MCP servers extend coding and productivity agents with instructions and scripts. Installing one gives its author influence over what your agent reads, writes and runs. The risks we found reviewing community skills, a review checklist, and the sandboxing and policy controls we use.
Users of our internal assistant kept repeating the same preferences: their team, report format, time zone. Adding memory sounds simple and is not. How we separated short-term and long-term memory, decided what to store, handled conflicts and stale facts, and gave users control over what the assistant remembers.
Teams wanted AI assistants to answer questions about internal ticket and order data through the Model Context Protocol. A local stdio server is easy; a shared remote one that respects each user's permissions is not. Our design for authentication, authorization, read-only tools, limits and audits.
Our internal assistant answered questions about supplier contracts and price lists, and its answers about prices were confidently wrong. Plain text extraction had flattened PDF tables into run-on text. How we found it, switched to layout-aware parsing with tables as chunks, and measured the fix.
A summarisation service trimmed documents to fit a token budget, yet some requests failed with a context length error. Our counter used a tokenizer from a different model family, ignored the chat template, tool schemas and the reserved output tokens. How we found the gap and switched to the provider's token counting.
A coding agent implemented a discount rule and wrote tests for it. Everything was green, the review looked fine, and production applied the discount to the wrong orders. The agent had written tests from its own implementation, so they asserted the bug. How we changed the workflow so humans own the acceptance criteria.
Our first voice assistant prototype took two to three seconds to answer, and callers talked over it. Most of the delay was not the language model. How we built a latency budget across audio transport, end-of-turn detection, speech recognition, the LLM and speech synthesis, and what we changed in each stage.