Sébastien Tang

I built an AI exec assistant that calls me out when I procrastinate and improves itself weekly. Just open-sourced it. github.com/sebastientang/…

Year two. Client hits scale. The data model we built can't flex. We over-engineered relationships for edge cases that never happened. Three weeks to refactor what took three days to design. Scope for what exists. Not what might.

Seen this pattern across a dozen small Salesforce teams. The ones that ship cleanly don't have better tools. They have fewer steps, enforced consistently. Simplify to what your actual team will follow. That's the release process.

Step 6: Freeze windows matter more at small scale, not less. Big teams have rollback muscle. Small teams don't. No deploys Friday afternoon. No deploys before a product demo. The blast radius of a bad deploy is proportionally larger when you're three people.

A 3-person Salesforce team with no Git discipline will ship broken code to prod. Not because they're bad. Because the process assumes a team that doesn't exist. Here's what actually works.

SI pricing model: sell days, bill days, repeat. The client pays for presence, not outcomes. Every architect in the room knows the last 3 sprints were padding. Nobody says it out loud.

Week one, I don't build anything. I read the org. Audit the data model. Map the blast radius of every active automation. Most clients expect output. I deliver a risk map. That's the difference. Profile if you need one.

Governor limits aren't a Salesforce quirk to work around. They're a forcing function toward good architecture. If your design fights them, the design is wrong. Full thread above. More in bio.

Step 4: Build a limit budget per transaction type during solution design. For each key transaction: estimate SOQL, DML, CPU, heap. Compare against limits before you write a line of code. I've done this on paper in 30 minutes. It's caught 3 rearchitects before dev started.

Most orgs don't design a governor limit strategy. They wait until a batch job fails at 2am, then fix the symptom. Here's how to design the strategy before the incident.

Second client. Same tool. Different org. Same three findings: dead automations, orphaned fields, and a Flow that nobody owns but everyone fears. Org debt has a pattern.

Agentforce adoption isn't blocked by the product. It's blocked by dirty CRM data, missing relationships, and schemas that made sense in 2018. You can't agent your way out of a data model problem.

Seoul taught me deadlines are a social contract, not a calendar entry. In Paris, slipping a date is negotiable. Here, it's a signal about who you are. 14 years in consulting. That reframe changed how I run every project.