As a product manager, I have created several agents to delegate areas of my work. I have one for backlog grooming, one responsible for maintaining metrics and analytics and one for design reviews. The recent Claude Code upgrades have made this possible.

I have to admit… AI has made Jira finally bearable…

Setting agent boundaries is delicate… like setting software boundaries… what should you allow your agent to do and what should you assign to another agent. When will agent complexity start becoming a thing…

Having fun building a Backlog Grooming agent… Main observation is: The core of the agent is a git repo. The agent writes what it learns from existing Jiras in its context in the repo, and also logs every action it takes in the repo. Observing the commit history is observing the agent working.

Apparently, the sexiest setup right for vibe coding is Claude code integrated in Cursor’s terminal! Agent inception.

Thoughs on Cloudflare’s MCP Code Mode announcement https://blog.cloudflare.com/code-mode-mcp/ There is a balance among the responsibilities of the different parties involved in an MCP interaction. LLM Host/Agent Provider Protocol User/customer It is true that currently, everyone is offloading costs to the user through tokens, and this causes a problem. However, I am not sure that shifting this completely to the provider is the right choice, at least for now. In the Code Mode LLM user tool context, the context is minimized, but code generation, the middle layer still consumes tokens, and I am not seeing a comparison between the size of the API and the size of the generated code.

Here’s my process for using questions to navigate ambiguity: Start with a “question dump” Organize by power Seek answers systematically Iterate and refine Document and share https://alifeengineered.substack.com/p/the-art-of-dealing-with-ambiguity

Every week is 2% of a year

In the US, being soft-spoken, deferential, and respectful makes you a valued colleague, NOT a leader. America was born in revolution, has an enduring cowboy identity, and today is led by a reality TV personality. This pattern plays out in business, too. US business culture favors those who speak quickly, act boldly, and ignore at least some rules and limitations. If anything, this trend is increasing with ideas like “founder mode” getting broad attention and support

Using natural language (or diagrams) to describe how to break down and precisely solve a novel problem (scalably and securely) is not easy and never will be easy. The key skills needed for software engineering are: requirements gathering / analysis, abstraction, data modelling / ontology building, algorithms, modularity / separation of concerns, team building / communication, threat modelling / security, debugging / problem determination, operations and monitoring, and ultimately… Not Building The Wrong Thing.