Ola. Ola.

Product Manager by day | AI-assisted developer at night

[Ola. Ola. @bishopdotun]

What makes OpenDevBrowser different?

- Supports Base64 (Encodes screenshot as binary‑to‑text ASCII characters like A–Z, a–z, 0–9, +, /.)
- Supports CLI + CDP + Extension + UI annotation

"npx opendevbrowser"
"npm install -g opendevbrowser"
To start: "npx opendevbrowser serve"

[Ola. Ola. @bishopdotun]

I released an update to opendevbrowser. Give it a try.

https://t.co/BKLVPjE5Rx

[unusual_whales @unusual_whales]

BREAKING: Meta, $META, took a $355 million tax break on Mark Zuckerberg’s $4 billion compensation by classifying him as a researcher, per the NYT

[Aravind Srinivas @AravSrinivas]

We're opening access to anyone (no Perplexity account needed), and we will run all tasks delegated via email for free for a limited time. Forward or CC [email protected] to get started, and the agent will complete the task in the background while keeping the email context.

[Ola. Ola. @bishopdotun]

@iannuttall I tried this with opendevbrowser. It’s such a difficult task. It works sometimes but still a hit or miss. Try the /shopping feature. It’s a cli and it’s deterministic. I added some big retail sites as providers already. https://t.co/BKLVPjDy1Z

[Ola. Ola. @bishopdotun]

@pvncher 5days. I tend to start a new one when I finish with a line of task on a feature. But I feel sometimes I have to steer. It sometimes gets off track. Seen that a few times with xhigh. It just gets stuck in a loop or starts talking about a task done a few compactions before.

[Ola. Ola. @bishopdotun]

@oldstackjournal OpenDevBrowser is a fully-loaded browser automation runtime. /research, /inspiredesign, /shopping, /computeruse etc. USP is token-efficiency. Tools, Cli & Chrome Extension.

https://t.co/BKLVPjDy1Z

[Ola. Ola. @bishopdotun]

https://t.co/BKLVPjDy1Z

[Ola. Ola. @bishopdotun]

Orchestrating agents in a loop shouldn't be complicated. One Orchestrator, many threads/sub-agents.

In this video, I'm making changes to the /InspiredDesign tool to improve how it extracts insights from video & GIF media for motion design using FFmpeg & FFprobe under the hood.

[Ola. Ola. @bishopdotun]

OpenDevBrowser v0.0.37

The new release has a fully implemented inspiredesign workflow. It's able to extract design inspiration from the internet (& Pinterest) for anything you're trying to build, whether you're trying to build a prototype, a component, or any part of your UI.

[Ola. Ola. @bishopdotun]

@thdxr @RhysSullivan 🤣🤣

[Ola. Ola. @bishopdotun]

@neetcode1 I see so much hype and BS but when you put the hype to test, the models fall apart. My recommendation is to scope your tasks and drip-feed the models.

This is how I setup:

Orchestrator <> Manager <> Validator <> Executor <> Explorer

The orchestrator gets the scoped plan.

[Ola. Ola. @bishopdotun]

Hot take on GLM-5.2. It needs a "baby-sitter".

Impressive with designs. Excels at focused single-threaded tasks.

Lazy with tools. Would rather brute-think everything alone. Struggles with complex workflows. Needs a bigger model to delegate & audit it's work.

[Ola. Ola. @bishopdotun]

Agents are good with one-shots but bad at iterating designs. Opendevbrowser cli solves that via "/inspiredesign & /harvest" feature. Uses ffmpeg & ffprobe under the hood to extract design inspiration from the web & media.

"npm install -g opendevbrowser"

https://t.co/kk5QAizcx9

[Ola. Ola. @bishopdotun]

@jxnlco There should be a 3rd option: Update. To provide additional information to the agent while still working without destructing the current work or changing direction.

Queue waits until current task is done. Steer changes the direction of operations mostly which isn’t optimal.

[Ola. Ola. @bishopdotun]

@ryancarson If you want your code factory to be independent of the labs and you are not tied to one lab and you are not switching back and forth, which provider do you use?

Devin? Cursor? Even with that, you are still using either of the 2 frontier labs (Openai/Anthropic) models.

[Ola. Ola. @bishopdotun]

@theo The $40 plan is no longer fit for purpose. $40 would be gone in about 1day of active dev time.

[Ola. Ola. @bishopdotun]

@theo The $40 plan is no longer fit for purpose. $40 would be gone in about 1day of active dev time.

[Ola. Ola. @bishopdotun]

@OpenAI how can I change the phone number attached to my chatgpt account? I created my account longtime ago and I no longer have access to the phone number. Now Codex is sending verification to that number's whatsapp? Getting access is almost a nightmare right now.

help pls..

[Ola. Ola. @bishopdotun]

@pvncher @OpenAI @romainhuet @RepoPrompt Congratulations @pvncher

Well deserved. Looking forward to what you ship next with the @OpenAI

[Ola. Ola. @bishopdotun]

@jlongster Since the frontier harnesses and agents are optimizing for long running sessions, how does staying in the loop work here? Except you are assuming that staying in the loop means reviewing after every session. But that’s almost unsustainable without agents?