Working at an agency brings about a smorgasborg of random projects at any one time. My list right now includes a proof of concept build for an agentic tool, a website that needs turned around by Monday, automating SSL certificates, resolving a DMARC issue on a mailbox. Then throughout the day I'll add random asides on like migrating a WordPress plugin functionality to Hubspot or some other random idea that strikes my fancy. What can I say I'm a bit code-
optimistic at the moment.
It is a lot of context to keep organized and moving forward at any given time.
The way I want AI to work for me is to get all the context out of my head and dished out to the various agents at my disposal to get things started in the right direction until I'm ready to take over or nudge it in a new direction.
I've experimented here and there with a ChatGPT agent that runs a weekly standup, summarizes the updates, and spins up a deck for a meeting. Or another agent that summarizes ticket status, recaps Slack channels, and pulls in client news for opportunities. But those still have mostly required me to be at keyboard.
I'm intrigued by the latest updates to ChatGPT and Claude's voice modes. Letting their agents call tools and spawn background agents to use the latest models to spawn off work. But the process I still want to do is use voice to build my to-do list and then spread it around to Claude, Codex, or Copilot to make sure I'm leveraging my token budgets to the max.
I want to start my morning off with a brain dump of action items that I can send out to agents and be able to hit the ground running when I pull into the office after my commute. Or be able to spin-off an idea for an proof of concept and get an agent working on it while I'm wrapping up the work that pays the bills.
Hell, I even bought wired headphones due to the latency trying to use voice to text dictation with my AirPods and macOS dictation apps like Handy. We'll see if that helps or if its just that I mumble too much.
There's more that will need to happen to build out this process, but I feel like we're inching closer. Now its just finding or building the tools that get me there.
🎙️ Model
Introducing GPT-Live
OpenAI’s latest voice mode brings tool use into voice mode. It does so by allowing you to kick tasks off in the background to other models while GPT-Live keeps responding to you with to create a better conversation feel. In my testing I was able to cover multiple projects and have ChatGPT trigger a background summary of one topic while we moved onto the next.
Not to be outdone Anthropic made their own improvements to their voice mode letting it use their Opus and Sonnet models alongside Haiku. It gives the voice mode tool use as well as allowing it to function in many more languages.
🔓 App
Handy
The open-source equivalent to Whispr Flow. Local models are getting more and more powerful handling voice to text and clean-up on device. Handy is a nice alternative to paid services like Whispr Flow if you're looking to test out and build the habit of talking over typing. As long as you don't mind talking to your computer in an open office setting.
🔌 Protocol
WebMCP: give agents tools, not DOM
WebMCP is starting to gain steam with ChatGPT recently adding support. It opens up a new framework for agents to more effectively interact with web applications by exposing the tools an agent can use in a way that’s easy to discover and understand. It should simplify agents discovery of functionality on a site and replace needing to scrape the site and interact with it via it’s own browser.
Good follow on WebMCP: Sarah Drasner (@sarah_edo)
🤖 Bots
Grok Bot
Grok Bot looks to give you a familiar UI (think iMessage) for working with agents. It gives each agent its own personality and sandbox computer while also allowing them to be connected and interact with each other. It’s a use case that was possible with OpenClaw with a lot of tinkering and setup, but the SpaceXAI team is trying to make more accessible.
Thanks for reading. Now go talk to your computer outside.
