Domain Methods

The safest strategy in business is often the riskiest one. Hear me out. Risk aversion feels responsible. It often sounds like: “Let’s wait until we have more data.” “Let’s not disrupt what’s working.” “Let’s avoid making a mistake.” “Let’s revisit this next quarter.” “If it ain't broke, don't fix it.” But over time, excessive caution becomes its own form of underperformance. The problem isn’t that businesses should take reckless bets. They shouldn’t. The problem is that many companies confuse risk management with risk avoidance. And those are not the same thing. Risk management asks: “What could go wrong, and how do we design around it?” Risk avoidance asks: “What decision keeps us safest from blame?” That difference matters. When teams become too focused on protecting the status quo, they often delay the very moves that create future growth: investing in new capabilities, testing new business models, adopting better technology, entering new markets, or challenging legacy assumptions. The end result is rarely dramatic at first. It generally looks like slower decisions. Smaller bets. More internal consensus. More analysis. Less experimentation. Then, one day, the business realizes it's behind its competitors, not because of a single bad decision. It underperformed because of a long series of safe ones. The best organizations don’t eliminate risk. They build the discipline to take smarter risks: Experiment in small increments. Measure what matters. Create decision rules. Learn quickly. Scale what works. Stop what doesn’t. In a changing market, standing still is not neutral. It’s a bet. A bet that clearer information will arrive before a decision has to be made. But often, that information either arrives too late or never arrives at all. And when a decision can be partially or fully reversed, waiting may carry more risk than moving forward. The question isn’t whether your business is taking risks. The question is whether it is taking the right ones.

Anyone who used the Databricks Assistant in the past knows it wasn't very helpful, or competent... or anything. Well, it's been "fired," and its replacement is actually good. Welcome, "Genie Code". The new 'boss' is NOT the same as the 'old boss'. Built into the Databricks Workspace, and it's completely free to use. Unlike the old Assistant, which operated more like a chatbot, Genie Code can do more and act more agentically. It even supports augmentation using Skills. (and MCPs, but who uses those anymore?) In fact, one of my internal connections told me they hardly do any "demo prework" anymore because the new Genie can basically do it all on the fly in front of the customer without any issues. Quite the difference from before. And they aren't skimping on the AI either. Rumor is they are using a frontier AI model behind the scenes. And you can tell. It seems to be on par with the latest and greatest LLMs we all have access to elsewhere. The only downside I've seen is that the new Genie is available only within a Databricks workspace. This means if you are more of a dev and live more comfortably in your terminal and/or VS Code, then this probably won't be as useful to you. But next time you are troubleshooting a job in a workspace, building a complex SQL statement, or trying to create a Databricks dashboard, find the familiar icon in the upper right-hand corner and give it a shot. I'd be surprised if it doesn't become a part of your toolbox, at least in-platform. 💫

Every Monday, they rebuilt the revenue report from scratch. Not because the dashboard was broken. Because nobody trusted it. 3 hours. Salesforce export. HubSpot export. Finance spreadsheet. VLOOKUP to stitch them together. Then 45 minutes explaining the delta to the VP. It was a $30M SaaS company. They'd spent $80K on a BI tool the year before. Still doing it by hand. I almost prescribed a new tool. Two weeks in before I found the Monday spreadsheet. Not auditing the trust gap first was the REAL mistake. The $80K was fine. What they bought worked. Nobody had fixed why the team stopped using it. That's not a tools problem. That's a maturity problem. Four stages from where most ops teams are to where they think they are: → Chaotic: Reports built by hand. No single source of truth. Data lives in 12 places. → Reactive: BI tool deployed. Half the team uses it. Half still emails the spreadsheet. → Proactive: One source. Decisions wait for data. No one fights over the number. → Predictive: Modeling what's coming. Not explaining what happened last week. Most ops teams think they're Reactive. They're Chaotic with better-looking spreadsheets. The dangerous part: the manual workaround becomes the process. The person who rebuilds it by hand becomes the single point of failure. One resignation away from losing the report entirely. 3 weeks. One pipeline: Salesforce + HubSpot + finance feed. One number. No delta. Everything well-documented so the next person who touches it doesn't need to call me. First thing I map when I start with a new ops team: what's being rebuilt manually every week that shouldn't be? I ask the person doing the rebuild, not their manager. There's always one. What report is your team still rebuilding by hand because they stopped trusting the automated version? Or is Monday morning still "spreadsheet time"?

Everyone is talking about what AI can do. Fewer are talking about if these LLM-driven workflows are dependable enough to run a business. That is the real shift happening in AI right now. In a demo, almost any "mouse trap" can look impressive. In production, the questions get harder: - Will it give a usable answers consistently? - Will it stay grounded in the right data? - Will it know when not to answer? - Will people trust it enough to act on it? That is where most AI projects fail. The trick is not generating usable output. The trick is generating output that is consistent and trustworthy enough to support decisions. Some might say, "yes, but AI will be directionally right most of the time". For teams in revenue, marketing, and operations, “mostly right” still becomes risky when applied to forecasts, attribution, lead routing, prospect messaging, or executive reporting. So I say we need: Less prompt theater. More grounding. More evaluation. More governance. More design for trust. The teams that win with AI will not be the ones with the flashiest demo. They will be the ones whose utility people can actually trust.

Maintaining column descriptions in your data warehouse used to be one of those tasks that was important enough to feel guilty about skipping, but tedious enough that you skipped it anyway. Now it's a 20-minute job: 1. Connect Claude to a read-only client for your warehouse 2. Have it pull a random sample of ~100 rows per column across your tables 3. Let it infer and suggest column descriptions based on actual data 4. Generate a single SQL file of ALTER statements to apply them 5. Review the output. Run the file. Profit... err... Done. Every table in your warehouse now has up-to-date human-readable column descriptions. What a world we live in.