Build in public

BlogWhat I write
after the check.

One question per article: what should get built, and how I would know it worked.

Articles

Topics

Latest

Oct 19

Browse articles

← Back to home

All articles

Filter by topic or search titles and tags. Each post answers one question.

DeliveryOct 19, 20264 min

The agent can finish it. That is not the decision.

Agentic coding removed the cost that used to kill a weak idea before anyone built it. I keep a refusal that can win, and an assistant does not move a finished draft onto the list of things I will stand behind.

DeliveryAgentsLearning

Read

LearningOct 12, 20264 min

When the tools build the report, write the contract first

I build reports with the Power BI MCP and reporting skills, so an assistant can produce the model and the pages. The contract has to state how the analytics are allowed to behave before that build, because the unit is the report, not one measure.

LearningAnalyticsPower BI

Read

DeliveryOct 5, 20265 min

Before an agent writes back, name who is allowed to speak

Barcelona's Copilot and Power BI updates let a report's definitions leave the report and change stored data. I would name the speaking model, the allowed writes, and who can stop a bad action before I turn that on.

DeliveryAgentsAnalytics

Read

LearningSep 28, 20265 min

When AI can build the report, the brief becomes the work

As report construction gets cheaper, I need a testable brief and a separate check for decision quality before a polished page earns trust.

LearningAnalyticsReport authoring

Read

DeliverySep 21, 20265 min

When access rules become cheap, policy becomes the bottleneck

AI can remove the repetitive work of configuring data access, but I still need a desired-state policy, a readable diff, and a named owner before permissions change.

DeliveryAnalyticsOneLake

Read

DataSep 14, 20265 min

When a report slows down, memory is not a change log

I treat semantic-model modification times as incident triage: useful for narrowing suspects, never enough to prove causality.

Power BIAnalyticsIncident response

Read

DeliverySep 7, 20266 min

A platform update should delete work, not add another path

When an analytics platform moves, I adopt the change only if it removes a named recurring job without removing the control that kept that job safe.

DeliveryAnalyticsData platforms

Read

DeliverySep 2, 20266 min

What one operator can build when projects stop being islands

Shipping many side projects teaches almost nothing if each one starts from a blank slate. Here is the shared spine I reuse: write things down, share knowledge, remember across work, and keep a human in the approval loop.

DeliveryPersonal opsPortfolio

Read

LearningSep 2, 20266 min

When the coach got gate ranks

My training plan already had honest numbers. They still felt like a parking ticket. Gate ranks wrap the same sensor data in a story so I actually look at readiness before I look at Tuesday squats.

TrainingPersonal opsHabits

Read

DeliverySep 2, 20267 min

When the portfolio needs a memory

Chat history is a transcript, not a filing system. After retiring a heavy local orchestrator, I built a quieter portfolio memory so each new session starts with what I already decided.

DeliveryPersonal opsPortfolio

Read

DataAug 31, 20265 min

When the reporting stack moves, preserve the contract

A native Power BI feature can remove a workaround, but the durable improvement is deciding which responsibility moved and what the report can still rely on.

Power BIAnalyticsFabric

Read

DeliveryAug 24, 20264 min

Agents cannot smell bad data. Start there.

Sadalage and Chandrasekaran argue agentic AI fails on sand foundations. As a delivery lead I would fix contracts and quarantine before another orchestration demo.

AgentsAIData

Read

DataAug 17, 20264 min

The agent can count delays. It still cannot cancel the flight.

Mohammad Ali's usable split: a semantic model answers a mathematical question; an ontology is for the process that acts. I would not skip the middle doll because Desktop grew a prompt.

Power BIAnalyticsAI

Read

LearningAug 11, 20267 min

Why Mondays stopped negotiating with me

A personal coach that turns Intervals form, Slack RPE, and block research into a Sunday week plan, so I spend energy training instead of re-deciding the week.

TrainingIntervalsSlack

Read

DataAug 10, 20265 min

When the report becomes an app, the semantic model becomes the contract

A React frontend can change the reporting experience quickly, but durable analytics still depends on a stable semantic contract and two deliberate release lanes.

Power BIAnalyticsReact

Read

DataAug 3, 20265 min

When an agent can change Fabric, the review boundary becomes the product

MCP and CLI can move Fabric work from clicks to agent actions, but the durable gain comes from scoped permissions, inspectable plans, and deliberate approval.

FabricMCPAI Agents

Read

DataJul 27, 20265 min

When analytics becomes an app, the semantic model is no longer the finish line

Fabric Apps move analytics beyond the report canvas, but a custom frontend only pays off when the experience needs more than a report and the team accepts software ownership.

Power BIFabricAnalytics

Read

DataJul 26, 20263 min

Nordic Equity Heatmap — when Power BI treemaps stop being enough

A Vite board for Nordic large-caps: market-cap size, day-change color, logos, and sector zoom — fed from the same gold the Power BI report uses.

Power BIReactData Viz

Read

CareerJul 24, 20264 min

From risk analyst to delivery lead — what actually changed

What shifts when you move from PD models and ECL to leading Azure application operations, and which quantitative habits still decide the call.

CareerDeliveryAzure

Read

DeliveryJul 24, 20265 min

When AI accelerates work, prioritization becomes the job

Individual copilots make people faster; without shared judgment and a real priority gate, that speed mostly manufactures misalignment.

AgileAIDelivery

Read

DataJul 23, 20264 min

Ledger — a research loop so I stop adjusting stocks on gut feel

A small Python research loop that compares a few stock strategies on shared data, changes the working rule only when the edge is clear, then lists concrete buys and sells with reasons.

PythonResearchEquity

Read

DataJul 23, 20264 min

Power BI portfolio — Nordic Boardroom reports that survive a cold open

A shared report grammar for analytics work: gold tables, a semantic model, Landing → Pulse → Drivers → Queue, then screenshots synced into Orbit so the site matches the repo.

Power BIFabricAnalytics

Read

DataJul 20, 20264 min

A Teams ping is not ops: what Fabric Operations Agents still need from you

Chris Webb proved Fabric can alert on a broken DAX visual. The harder work is writing semantic instructions, validating KQL, and deciding what the owner does when Teams rings.

Power BIFabricAgents

Read

DataJul 20, 20264 min

Direct Lake isn't free: when Fabric's cool mode earns its keep

Chris Webb's Direct Lake case is really about where cost and cycle time live. For portfolio reports that still ship as Import models, measure the gold layer before you chase framing.

Power BIFabricDirect Lake

Read

AIJul 13, 20264 min

Useful work per dollar — what delivery and analytics should change first

If agents become a unit of work, stop counting runs and start defining useful outcomes — the same discipline that makes a Monday dashboard trustworthy.

AgentsAIDelivery

Read

AIMar 28, 20265 min

Building Orbit — the personal HQ I actually maintain

A Next.js headquarters for Blog, portfolio proof, a curated RSS feed, and Studio — designed so future-me can ship without renting another platform.

Next.jsPortfolioAutomation

Read

BlogWhat I writeafter the check.

All articles

The agent can finish it. That is not the decision.

When the tools build the report, write the contract first

Before an agent writes back, name who is allowed to speak

When AI can build the report, the brief becomes the work

When access rules become cheap, policy becomes the bottleneck

When a report slows down, memory is not a change log

A platform update should delete work, not add another path

What one operator can build when projects stop being islands

When the coach got gate ranks

When the portfolio needs a memory

When the reporting stack moves, preserve the contract

Agents cannot smell bad data. Start there.

The agent can count delays. It still cannot cancel the flight.

Why Mondays stopped negotiating with me

When the report becomes an app, the semantic model becomes the contract

When an agent can change Fabric, the review boundary becomes the product

When analytics becomes an app, the semantic model is no longer the finish line

Nordic Equity Heatmap — when Power BI treemaps stop being enough

From risk analyst to delivery lead — what actually changed

When AI accelerates work, prioritization becomes the job

Ledger — a research loop so I stop adjusting stocks on gut feel

Power BI portfolio — Nordic Boardroom reports that survive a cold open

A Teams ping is not ops: what Fabric Operations Agents still need from you

Direct Lake isn't free: when Fabric's cool mode earns its keep

Useful work per dollar — what delivery and analytics should change first

Building Orbit — the personal HQ I actually maintain

BlogWhat I write
after the check.