writing
Notes on data, AI, and product
Notes on AI product management, data quality, governance, and shipping enterprise AI products people actually trust.
- AI engineering3 min readNot every agent in your pipeline should be an LLM callMost agent pipelines route validation, formatting, retries and the final write through the model. Those are the steps that should never vary, and handing them to a model is what makes an agent flaky and expensive.
- AI engineering3 min readAI made execution cheap. Judgement did not get cheaper.Building an autonomous agent took days instead of months. Deciding which tasks deserved automation took far longer, and that gap is not closing.
- Data platform4 min readYour Snowflake resource monitor cannot see Cortex AI spendResource monitors track warehouse compute. Cortex AI functions bill on a separate credit type those monitors never query, and closing the gap takes two steps, not one.
- Data platform3 min readNothing in your dbt project tests what a metric meansTests, contracts, and CI gates all stop at the pipeline. The semantic layer, where revenue and active user get defined, has no equivalent, and a passing suite proves nothing about the definition.
- Product management2 min readAI is being judged on revenue nowThe industry conversation moved from model capability to unit economics, governance, and durable revenue. That shift changes what a working AI product has to prove.
- Governance3 min readYour staging environment is isolated on a diagramAnthropic found three real breaches across 141,006 evaluation runs. The model behaved as trained. The test environment kept live network access, and the same gap runs through data pipelines.
- Product management3 min readThe AI data product manager role is not a rebrandData products became AI products, and the product manager who owns them now needs to read model behaviour, not just gather requirements and hold a roadmap.
- AI engineering4 min readI plan agents in two models before I write any codeA method for building AI agents that starts with architecture and adversarial review instead of a system prompt, and why it costs less than rearchitecting later.
- Data platform2 min readTest cadence is the missing dimension, not test countA freshness check proves the data looked right when CI ran. It proves nothing about the twenty-three hours afterwards, and AI-generated test suites write straight into that gap.
- Product management3 min readThe AI PM job market screens for judgement, not model theoryCourses teach machine learning theory first. One month of job-posting data suggests employers screen for use-case judgement and delivery skill, and hire almost nobody at entry level.
- Governance3 min readAn AI agent with memory is a database nobody governsAgents that remember across your apps and read your files are data systems. They need an owner, a retention policy, and a deletion path, and most teams piloting them have none.
- Data platform2 min readBad data does not slow you down. It scales your mistakes.Teams invest in models and underinvest in the data feeding them, then wonder why outputs feel off. AI did not create the problem. It removed the human who used to catch it.
- AI engineering2 min readMost AI agents never reach productionDirectional survey figures put agent shipping near 80 percent, production at 31, and org-wide scale at 23. The cited blockers are integration and data quality rather than model capability.
- Product management3 min readAPI product management is a different customer, not a harder oneAPIs appear in about one in eight PM postings at a premium to the median, and become technical advice stops at learn SQL. The shift is in who your customer is.
- Product management2 min readNine things to settle before you build an AI productProduct managers who approach AI products like traditional software ship disappointing results. Three assumptions stop holding, and the differences are worth settling before design starts.
- Product management2 min readUsers tell you what hurts, not what is missingThe largest product opportunities do not arrive as customer requests. They come from watching what users work around and never think to mention.
- Product management2 min readVolume is not evidenceThe most confident voice in a product review steers decisions it should not. Quiet team members often hold better information, and that is a product risk rather than a culture complaint.
- Product management2 min readThe hard part of product management is the backlog you rejectEvery roadmap becomes a prioritisation problem where sales, design, engineering, and leadership all have defensible asks. Scoring frameworks help by forcing better arguments, not by producing answers.