From
could AI do this?
To
Something people use every day.
I'm Justin, a Lead Software Engineer at Salesforce in Singapore. I build it, decide what it's allowed to do, check that it's right, and keep it running.
Same four steps, both places
- 01
Build
At workNew platform capabilities taken from design to production: APIs, an MCP server, and the access model that lets agents act for a person.
At homeAn MCP server with narrow, named tools so Claude can answer questions about our finances.
- 02
Guard
At workSecurity and design questions raised at review time, and guardrails for how the team builds with AI.
At homeClaude's access is off by default, writes sit behind a second switch, and the database enforces the rules.
- 03
Measure
At workAutomated checks that decide what can merge, and a second look from me before a risky change deploys.
At homeAn eval ladder over 450 real transactions: 58% right where the model decided, 90% where memory did.
- 04
Operate
At workWhen something breaks I find the root cause, fix it, then prevent that kind of failure and write the runbook.
At homeAn AI reviewer hung for six hours in CI. Every "ask" became "deny", with a time limit taken from run history.
In daily use
Selected writing
All writing →Experience
More →- 2022 – now
Salesforce, Lead Software Engineer. Technical lead for the backend of an internal communications platform that reaches employees through agents in Slack. I build it, release it and keep it running.
- 2019 – 2022
Salesforce, Software Engineer to Senior. Co-led the order automation system that took Slack's order processing from 70% to 94% automated.
- 2017 – 2019
Workato, then Slack. Solutions engineer, seconded to Slack's IT team as an integration consultant, then an associate software engineer there.