Darren Burns

building delightful terminal tools
engineering @monzo

[Darren Burns @_darrenburns]

Posting 3 is almost there.

Should I stick with the classic purple default theme or switch it up for the new major version? 🤔

[Darren Burns @_darrenburns]

More bullish on @AmpCode by the minute.

It's supplying before/after screenshots, and inline interactive windows allowing you to check it's work right within the chat.

Here I am testing out a change to a TUI demo app inline.

[Darren Burns @_darrenburns]

Ok, @AmpCode Orbs are nice.

Asked Amp to fix the location of the IME popup in a TUI framework to ensure it tracks the cursor correctly, and it proved it's work by showing me a before/after image of the terminal with the OS-level IME popup appearing in the correct place.

[Darren Burns @_darrenburns]

Skills with CLI wrappers around the tasks they perform have been incredibly helpful for me, particularly when it comes to agentic verification and creating evidence that a task has been completed.

Without a CLI wrapper, I found the output lacked consistency - agents tried different approaches and the output always looked slightly different and used different tools each time. It likely also cost more tokens vs reading the —help for a simple focused CLI.

Another benefit: a CLI wrapper is easy to allowlist for people working in environments with controls around permissions - otherwise agents can end up installing and trying a variety of different tools to perform tasks, each of which need allowlisted.

I now have simple wrappers around Playwright, tools for controlling mobile devices, watching PRs for comments and review status, and around tmux and ffmpeg for capturing screenshots and videos, trimming/cropping them and converting to GIF, and attaching them to PR descriptions.

[Darren Burns @_darrenburns]

I had an upcoming price increase that prompted me to move. You can export your passwords in 2 secs from the menu and import into another password manager. Was done in 5-10 mins. I chose Bitwarden and it seems decent enough. Was putting it off cos I thought it’d be a slog, but it’s all automated nicely!

[Darren Burns @_darrenburns]

@CyrilDaSwan I agree!

[Darren Burns @_darrenburns]

@skirtle19 Yeah fair point - I was just meaning in terms of total income+benefits. Overall could be worse off or better off depending on their individual circumstances.

[Darren Burns @_darrenburns]

@JohnPretty2o2o @Goodbob888 I’m talking about *families* with a single earner, which is very common (stay at home parent looking after kids, carers, medical reasons).

I have no idea what point you’re trying to make with this scenario…

[Darren Burns @_darrenburns]

@JohnPretty2o2o @Goodbob888 Myself and the original poster are in Scotland, so tax bands are different.

The essence of my point still applies to the English bands though - a couple each earning £42k wouldn’t be thought of as “rich” in England?

[Darren Burns @_darrenburns]

I'm pretty sure you're right. I find myself slowly converging towards what you've described.

At first I had one agent per task, then a manager agent per project, then one manager overseeing all projects.

All questions get bubbled up to the manager so there's a single surface where I answer any questions. I try to give the managers agency to make decisions without me - I'd rather they make a small reversible decision, record it, and move on than get blocked waiting for me.

An optimal review surface is really what feels like is missing to me. It's really difficult to grasp the current state of the world by scrolling through a transcript full of comms between agents. I want to understand the state of the world *right now*, like a live summary.

Maybe no such thing can exist - the volume of code and decisions being produced is just unprecedented and I struggle to even reason about what I'd want to see on such a review surface. Perhaps a section per agent summarising the state of their task (what stage are they at? planning, building, etc), with key decisions made, diffs, and any other output artifacts linked?

[Darren Burns @_darrenburns]

@Joshua_WD yeah that's essentially what it is. you can leave comments on the diff and then ask the agent e.g. in Claude Code to respond.

or, you can ask the agent to comment on the diff for you.

it's basically a shared diff surface on which you and an agent can see and interact with.

[Darren Burns @_darrenburns]

really enjoying herdr so far and the fact that agents can control it. works great with a diff review tool i'm working on.

[Darren Burns @_darrenburns]

@thorstenball Well I very much enjoy and look forward reading your newsletter every week. Keep talking!

[Darren Burns @_darrenburns]

i usually use a browser based tool for compressing, but it always felt a bit slow

i got claude to make something to speed up my workflow and 10 mins later this appears in my status bar.

better workflow, and compresses way quicker. prompt was like 2 sentences too!

[Darren Burns @_darrenburns]

@alekdobrohotov @ClaudeDevs https://t.co/WzwsefJ06n

but I think the mods are just kind of mixed in with everything else - can't find a filter anywhere

[Darren Burns @_darrenburns]

Opus 5.5 one shotted a landing page in 10 mins for https://t.co/OoqnBpd8Qh that is 100x better than the current one that took me a week

it even added working keybinds, "jump mode" navigation and a command palette theme picker, just like the TUI app has, without prompting.