Setup
- Claude Code
- Anthropic Teams plan (work)
- Claude Pro Max 200 plan (personal)
- Codex CLI + Deep Research
- OpenAI Enterprise (work)
- ChatGPT Pro subscription (personal)
- OpenAI Enterprise (work)
- Visual Studio Code
Claude Code + Codex CLI
The November releases of Opus 4.5 and GPT-5.2 were an inflection point - for the first time, the models could autonomously handle the day-to-day problems I work on. I now run more Claude Code instances in parallel since I can trust each to make real progress on its own and end up waiting on them if I only have three tasks going at once.
Skills have been the biggest shift in my workflow. Progressive loading means skills can deliver the same guidance as AGENTS.md without as much context bloat. Investing in skills has paid compounding dividends, both in terms of agent autonomy and project maintenance.
The biggest example is my personal Claude Code plugin marketplace at tvishwanadha/skills. The reviewer framework is the centerpiece. It’s a modular review setup that lets agents check whether they actually accomplished what they set out to do. While it isn’t part of the framework itself, I use the tools in the framework to make project specific ci-review variants for GitHub Actions. My coworkers rely on them to cut human review time on PRs even when they aren’t running self-review locally.
Deep Research
Earlier in the year I was bouncing between Deep Research and ChatGPT Agent mode. Agent mode had newer reasoning and was easier to drive, but Deep Research’s reports were noticeably better. OpenAI has since theoretically closed that gap by refreshing Deep Research onto GPT-5.2. I haven’t had to reach for it since the update but it was long overdue and I’m hoping it’ll be good when I need it.
Visual Studio Code
It’s an editor. It works when I need to read things.