Tom Bradshaw

Quick demo of JrDev - my coding agent built entirely in Python (backend + Textual frontend) 🐍 Such a fun project to build! Free & Open Source Link below 👇

@simpsoka Can't wait to see what product this turns into 👀

You’ve seen me talk a lot recently about continuous ai. Whelp, I’m hiring a pm on my team to help figure out how to do that. It’s gonna be a fun one to build. If you’re interested, check out the link in Kelly’s post!

@willmcgugan But... "Python is slow"!

Not sure if this is worth shouting about. But Toad starts up quicker than *all* the other coding agent CLIs. "toad -a claude" gets to a prompt in ~500ms "claude" drops you to a prompt in ~1500ms The irony is, you are still talking to Claude either way. It is faster to start because Toad displays the prompt before the agent has finished its startup process. I suspect Claude's own CLI and possible all other agents do something network related at startup. And you don't get a prompt until that has completed. Can't claim to have put any thought in to designing it this way. This is a fortunate side-effect of running agents in their own process. But it is still somewhat useful to be able to start typing your prompt immediately.

@willmcgugan @MistralAI Wow that is awesome. I love using Textual 🚀

Textual downloads are going through the roof lately. I suspect the last spike is @MistralAI 's new CLI, written in Textual. Toad is getting popular, but I doubt it contributes to that many downloads. Yet. github.com/batrachianai/t…

Opus 4.5 passes my "who is presstab" test with flying colors - the most accurate I've seen so far. This is an empty context knowledge check: no web search, no conversation history. Impressive given how obscure these facts are.

@willmcgugan If it is beneficial fore the data storage to be human readable (like llm chat history, configs, etc) then definitely JSON.

How would you prefer an app stored its local data?

@BuiltByVibes @julesagent Quite the bill for a side project

@presstab_dev @julesagent Primarily data volume. It wasn't great to start but was fine for low traffic. Then I got 30k hits last month and a $120 cloudflare D1 bill. So standard unindexed was part of it. The next step is moving calculation and generation to a worker and then serve the data off R2.

Got a high bill on a demo project that got popular (30k/mo) with inefficient SQL use I'm a frontend guy, oops @julesagent is going to fix this I gave it screenshots of the database usage in cloudflare, and the queries that are causing the problem I asked for a few different approaches, since the problem is beyond caching We forgot indexes, which even @jackfriks implemented after scale And we need a few API changes Jules has bash access and env variables so it can run the wrangler d1 migrations for us The API calls are updated as is the test suite to match the new data structure We're also offloading the realtime SQL queries, but that's another big task and we've already asked this instantiation to do a lot So I ask for a prompt for step 2 You won't hurt its feelings The next session will run the commands to set up R2 to just serve the mostly static list Don't ask one session to do too much! You have a lot of tasks a day, use them

@simpsoka If you actually want to complete the items on your to do list, nothing is more motivating than pen and paper!

I use ai a lot for creating lists, like to do lists, or instructions. And I wish the chat apps would turn those into actual checklists so I can mark which items are done. Just putting this random feature request into the universe.

@itsakdev 3.0 behind the scenes?

FYI: Some % of Jules users will start seeing more detailed agent thinking and more frequent & detailed updates. Let us know if you have any feedback!

@travisennis @iannuttall Does exacto only use high quant?

@iannuttall glm 4.6 or kimi k2. use the exacto variants if you want good tool calling ability: openrouter.ai/models?q=exacto

which open router model should I use if I want to build a chatbot that: - follows basic rules + remembers them - has ability to run simple tool calls - has decent context window - is cheap as possible what would you recommend?

@travisennis They can be so good at certain tasks, but sometimes they really get tickled the wrong way and decide to go nuts with the length and scope of their output.

Anthropic models just do too much. Too many things that just don’t seem worth the trouble.

@joelcompiles I'll take the blue one

here in my garage… i’m going to be talking about creating products for creators with Yamaha

@itsakdev I can definitely point to quite a few!

@presstab_dev There’s a reason “who was your favorite teacher” is one of the standard security questions. Most people have a teacher who they can point to that made their life better. Such a noble career and I bet you’re crushing it

I've been quieter on X lately... honestly struggling to find the time to work on personal hobbies. Teaching Databases II has consumed most of my energy (3X weekly, in person, 2.5 hour classes!). Never realized how much prep goes into teaching when you genuinely care about student learning: - Creating engaging labs that reinforce concepts - Designing classes that blend lecture/whiteboarding with hands-on work and strategic YouTube videos. Attention spans are finite, and databases can be a brutal topic at 6pm - Grading assignments while providing meaningful feedback - Answering the constant stream of emails and Teams messages (definitely underestimated this one) - Studying edge cases to keep everything fresh in my mind It all adds up, but I'm genuinely enjoying it!

@manueljishi @deedydas I forgot the name. It was over linkedin and the person was removed, so the message history is gone. It was a bitbucket repo, with a task that had frontend, backend, and data sections to it and some tasks to complete. It was pretty strange, definitely set off red flags.

@presstab_dev @deedydas Hey Tom, this code looks really familiar, is it from an interview from a company called Belo Labs?

This software engineer was nearly hacked by a coding interview. I've never seen this attack channel before. Be careful out there, especially vibe coders.

@BuiltByVibes @julesagent This is cool. Nice video pace too. Thanks for using Jetbrains, I have a hard time reading other IDE's 😅

A slack command to a remote agent can make a first pass on a bug fix from a report or enable teams to make style and copy changes without engineering The @julesagent API is out and we're going to build a slack bot together so you can unlock this ability for your team #sponsored

@catalinmpit Plenty of free models on openrouter to connect with opencode

I cancelled all my AI subscriptions. Let's see how far I can get only with free plans. I use OpenCode with GitHub Pro, which I get for free since I'm an open source developer.

@Hesamation A real terminal with a package manager is a benefit. Really have to use WSL to get that on windows, and then you aren't really using windows at that point.

I switched from Windows to Mac recently. It has 0 developer benefits.

@therafikmatta @_philschmid Jules was built as a web agent. Just barely got CLI after people begged for 6 months

The Jules team is shipping relentlessly, giving you more ways to build with Gemini including an API and CLI. 🐙🐙 > Jules directly in your terminal with the new CLI. > Integrate Jules into using the new developer API. > Context management with file call outs and environment variable manager. > Memory feature to remember your preferences.