Gareth Paul Jones πŸ’™ (@gpj) header
Gareth Paul Jones πŸ’™ (@gpj) profile picture

Product & Research Lead. Token Junkie. Formerly of @twitter, @apple, @google. Proud husband to @kristineyjones πŸ‘«

null
Joined July 2008
Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)Photo by Gareth Paul Jones πŸ’™ (@gpj)
I think OpenAI’s Decisions API is totally underrated. I’ve been building a bunch of small apps with it, and it's been a lot of fun. There are a ton of classification style solutions that were before too cumbersome to build and now it's just way easier. Here are some that I've been playing around with: 1. An ADHD Detector I watch far too much YouTube. This app takes a screenshot every couple of minutes and checks whether YouTube is visible or other things and alerts me that I am distracted. 2. An email triage app My inbox is out of control. I built an app that checks whether a message needs action, puts it in a category, and shows a spam score from 0–10. The app then uses that score to decide what to keep and what goes to trash. 3. A transaction categorizer This takes transaction descriptions and groups them into household, food, health and other expenses. Then the app adds up the categories. Merchant descriptions can be a mess, so having help with the sorting is pretty useful. 4. Gamer I can play tic-tac-toe or chess against the model. I’ve also been using it to experiment with decisions driving browser actions. 5. A notification router I have a notification pipeline, and I wanted the urgency of an event to affect how it reaches my phone. A routine update can wait. I like being able to try these ideas without first collecting a dataset and training a separate classifier. There are a bunch of places in everyday apps where a category, a score or a yes/no result gives you something useful to build around. Checkout the video for more..
2
1
19
740
Gareth Paul Jones πŸ’™ retweeted
It’s almost incomprehensible how wide the distance is between someone who’s working at the frontier with AI and basically everyone else right now. The crazy thing is these lines are not converging, but only growing farther apart. If the curve of AI progress had leveled off, it would certainly have given most people and companies time to catch up or implement their workflows in a standardized way. The technology would have neutralized competitive advantage because everyone would be using it the same way. But what’s happened is the opposite. The curve of AI progress is basically going vertical right now, which means if you had codified your processes 3-6 months ago you’re likely already behind again. Now imagine most companies that kicked into gear a year ago and thought that they were done. Even further behind now. I’ve never seen a technology in my life and career that improves at this rate, which consequently means you need to upgrade your expectations at a similar rate. Our brains weren’t wired for this, and our organizational change management processes certainly weren’t either. With any form of recursive self improvement, things just get wilder. It’s a time when you will need to be constantly reevaluating the things that didn’t work just a couple months ago to see if they now can be solved. AI native companies certainly have a strong advantage due to this right now, so now it comes down to can the bigger companies keep up. Wild times.
872
1,172
492
9,750
3,207,273
Foursquare was acquired last month. The outcome was disappointing. It's taken me a while to write this, but I think I'm finally done grieving it. Here are some thoughts on our final years, an effort to honor the team who helped build & maintain FSQ's legacy, and my attempt to finally get some closure. teendrama.substack.com/p/som…
444
269
324
3,007
738,528
Gareth Paul Jones πŸ’™ retweeted
Vibe coded my parents' TV this afternoon! It was showing a sponsored ad before you could even get to Netflix. Gave Claude Code the model number, it found the TV on wifi, removed 25 Xiaomi packages and wrote a new launcher. Looks like an Apple TV now lol
971
1,701
837
35,927
8,777,621
Gareth Paul Jones πŸ’™ retweeted
If you want a yolo setup for computer use
Replying to @BenjaminBadejo
defaults write -g ComputerUseAllowForbiddenTargets -bool YES
14
6
1
308
27,410
i named my dot "kitt". work, without opening a laptop. 1. commute. i call it. it remembers the follow-ups. 2. codex. it picks up a thread on my mac and nudges it along. 3. life. weekend plans with my daughter. a nas comparison i can read later. 4. reminders. birthdays. call this person. check this slack channel, ping me only if i need to act. 5. the recurring check. a channel, a pull request, a launch. same rule. ping me only when it matters.
1
7
423
my @bot just called a hotel with @usebland and added some room preferences for an upcoming trip asked some questions about parking. next: call @FrontierCorp πŸ‘€
1
163
jev-1 from @typesafeai is a new and interesting model and it's lightening fast <150ms. it provides typed choice, probabilities and confidence out. people have been building some interesting stuff with it from agent web browsing, game playing, order booking, search, classification, binary decision making (spam/not-spam), multi-choice, scoring and more. i made this simple tic-tac-toe game where you can play against jev: jev-tac-toe-replica.gpj.work… it's fast, fun and cheap. jev is free on @vercel's ai gateway until the end of the week πŸš€
4
297
the agents api from openai is really nice. codex in a box and i think most people are sleeping on this. i've been using it a bunch and it removes a bunch of agent anxiety away from a bunch of tasks for me. 1. automated tasks (release notes, dream mode, meeting prep, todo updates) 2. code reviews (post merge latency stats, auto-flagging, skill updates) 3. reporting (posthog reviews, agent $, skill usage review). the biggest benefit is not having to run top or htop to monitor memory/cpu/energy use.
3
2
7
419
I’ve been experimenting with audio models recently using models from @ElevenLabs and @DeepgramAI. here are some ways to get the most out of the latest audio models: 1. Word-level transcription You get a transcript with start and end timestamps for individual words. I am terrible at saying a lot of β€œums” and β€œuhs” and β€œlikes”. so this is particularly useful for editing or just helping me reduce the occurance of this. Some other contenders are Deepgram, WhisperX and OpenAI audio models. 2. Sound effects There is something pretty novel about extending the sense of sound for things like β€œdings” when you add a task? The sound effect model is really nice at adding sounds for all sorts of things like notifications, game effects or when something finishes. 3. Music Music changes the mood and style of things. I like being able to generate music to match an idea or let the model orchestrate some music to go along with a certain vibe. 4. Voice isolation Sometimes, recording audio can be frustrating with that hum from a GPU or the fans running to keep you cool. Voice isolation models do a really great job at removing some of that sound, but also be careful as cleaner doesn’t always sound more natural. 5. Dubbing Being able to record in your own voice in another language without recording the whole thing again is next level. Like the beginning of the video in this post into Spanish. The speed at which these audio models are improving is super impressive, but wow are they fun and interesting to play with.
2
1
7
249
I’ve had more fun with GPT-6 Astra than I’ve had with a model in a long time. Here are 3 quick tips to get the most out of the model based on my own experience: 1. Use the model as an orchestrator feels more important than ever. For example you should instruct the model to act as a primary agent and that it's job is to plan, coordinate, and delegate rather than perform all tasks directly. This is really helpful for cost control too. 2. Review previous skills. If you have skills that are really detailed or have some custom CLIs for things, it's a strong enough model that it will figure it out without these things. For example, instead of having a CLI for Davinci Resolve to edit the video below, I just point it at the Davinci Resolve python scripting docs and it'll just figure it out. 3. Use seed strings (YES really). Generate a long, random alphanumeric string using a shell script. Define the design direction of my idea (color scheme, layout, typography, etc.) based on the string. Look beyond the surface for subpatterns, special numbers, anything that inspires you. Watch it cook some interesting designs. Over the weekend, I've been able to visualize electrical cables behind my walls (drill ready), fix the wifi (finally), re-designed my personal website, improved SEO, schedule my work week and tackle tasks that required too much back and forth with prior models to get a desired outcome. If you haven't checked it out you should!
3
201
GPT-6 Astra is a game changer for home renovation. Uploaded a floor plan with some measurements and help visualize an indicator for studs/electrical/water/gas lines.
7
290
Gareth Paul Jones πŸ’™ retweeted
GPT-6 Astra is now available to all Pro, Enterprise, and Business Premium users alongside the API. I’ve been using it to build apps, LEGO models, and 3D scenes. Here are five things that changed how I use Codex πŸ‘‡
49
92
40
1,035
659,188
Gareth Paul Jones πŸ’™ retweeted
I'm writing a guide to pstack! Here's part one.
265
751
245
8,040
1,305,819
I think Omarchy is the first OS I’ve used that actually feels built for the next era of personal computing. After a week or so it’s like living in the future @dhh and team cooked!! Also it’s so much fun to be able to change stuff on your machine that was previously locked down by @Apple. Linux Distros were always kinda hard to setup without going to some random forum to figure out why your driver for something didn’t work and now you can just use a model with a harness and it will work instantly. It’s also completely Open Source so the best models can easily debug things, unlike closed OS like @Apple and @Windows . Here are some things I noticed: - The window management is really nice and much easier to utilize screen space compared to Mac. You can configure it however you want with shortcuts, hot keys change it based on the time of day. - It’s wild to just have your OS change with a few prompts with an agent using a Mac feels like driving a stick shift with no power steering. - If you want to change the button on your Logitech MX Master Mouse or Elgato foot pedal. Ask codex and in 10 seconds it’s done. - Change the keyboard shortcuts to be Mac friendly, ask Codex and done. - It has all the apps you probably need Chrome, Codex, Grok Bot, Slack, Ghostty, Discord, OBS, Davinci etc. - Whenever i had a problem with something not working like my airpods, i could just ask a model and then it would figure it out and just work. Also it’s crazy fast, none of the Apple or Windows bloat. More hardware manufacturers need to ship with it pre-installed. πŸ‘€ @FrameworkPuter @Dell omarchy.org
4
435
Gareth Paul Jones πŸ’™ retweeted
Single best discussion ever recorded on coding with agents.
SpaceXAI engineer (ex-Cursor): "right now I'm running 10-20 GrokBot agents that automate 90% of my routine i have a Chief of Staff agent. He knows about all my other bots and manages everything" in a 50-minutes podcast, a SpaceXAI engineer showed how to build a team of agents that will work for you 24/7 worth more than a $500 course on agentic engineering watch today, then read how to build a Grok agents team from scratch in the article below
36
108
3
2,585
691,836
The best thing about Computer in Codex is having it deal with certs and all the other stuff to get an App on TestFlight with iTunes Connect and the Google Play Store.
2
174
Gareth Paul Jones πŸ’™ retweeted
you won’t find this on threads
3
14
1,389
Gareth Paul Jones πŸ’™ retweeted
Today, we’re launching Bland Speech v3 - The world's first Human Speech Engine. In @designarena’s Audio Realism benchmark, Speech v3 is the top model, outranking Elevenlabs, Grok, Cartesia, and OpenAI. Trained on over 100 million real, human conversations: businesses, developers, and creators can now create any audio they can imagine. To showcase this, we used 5 seconds of old footage to help James, a 49 year old father who recently suffered a stroke, get his voice back. Try it for free at bland.ai/speech Full length documentary below🧡
237
304
224
2,217
2,136,824