The stack

What I actually use.

A living list of the AI tools in my stack, sorted by the job they do. For each category: what I use now, what I am testing, and the one I would pick. It changes as the tools do.

Updated 2026-09-24

Everyday assistant

The models I talk to all day, and where most writing starts.
Using now
ChatGPT logo

ChatGPT

The first tab open, for most of the day.

Claude logo

Claude

Writing and thinking when the words matter.

Experimenting with

  • InstinctA newer assistant, under evaluation.

Pick: ChatGPT. The most useful everyday default right now.

Research & analysis

Deep research, reading and data.
Using now
Claude logo

Claude

Research and analysis, often in Cowork.

Experimenting with

  • PerplexityFast, sourced answers from the web.

Pick: Claude. Best for reasoning over messy material.

Design & brand

Images, decks and visual identity.
Using now
ChatGPT logo

ChatGPT

Images and visual ideas from a prompt.

Claude Design logo

Claude Design

Corporate and generic design, on brand.

Pick: ChatGPT. The strongest image output for the least effort.

Building & code

Apps, prototypes and internal tools.
Using now
Claude Code logo

Claude Code

Plain English in, working software out.

Experimenting with

  • CodexOpenAI's coding agent, closing the gap.

Pick: Claude Code and Codex. Both earn a place. I run them side by side.

Agentic workflow

Agents that run tasks end to end, not just answer questions.
Using now

Hermes Agent

The agent behind my customer and ops workflows.

Experimenting with

  • GrokbotxAI's agent, under evaluation.
  • InstinctAlso in the running here, under evaluation.

Meetings & notes

Capture, transcribe and summarise.
Using now

Wispr Flow

Talk instead of type, everywhere.

Pick: Wispr Flow. Dictation that keeps up with thinking.

Voice & video

Video generation. Nothing has earned a permanent place yet.
Using now

Nothing settled yet. Still testing.

Experimenting with

  • KlingVideo generation. The current front runner.
  • SeedanceByteDance's video model, run through Dreamina.
  • Gemini OmniGoogle's video generation, under evaluation.
  • HiggsfieldGood output, but far too expensive to run.

Pick: Kling. The best balance of quality and cost for video.