Cursors killer feature, GPT 5.4 and Replit has something up their sleeve...

Lets dive in...

the headlines

Cursor released a new automations tool
Replit teases something spicy
OpenAI releases their latest model

Cursor releases "Automations" feature

Last night, I recieved an email titled "Introducing Cursor Automations" and, being the avid Cursor fan that I am, my heart skipped a beat.

Automations allow users to set up agents that work autonomously on a task once triggered, whether that is through a Slack message, timer, or GitHub repo push.

The Cursor team said Automations are particularly useful for:

Review and monitoring, where they can catch and fix everything from style nits and inconsistencies to security vulnerabilities and performance regressions.
Chores, where they can help with everyday tasks and knowledge work like triaging bug reports, summarizing code changes, and adding test coverage.

Claude Code users... is this enough to make you want to switch to Cursor?

Replit's CEO cannot get enough praise

The founder of Y Combinator, Paul Graham, has been praising Amjad Masad left, right, and center this weekend in two viral tweets.

The first tweet sent shockwaves around AI spaces, leaving users to speculate what Replit might be cooking up: vibe coding hardware, Replit's own model, or drag-and-drop visual editing.

Paul Graham then followed up with a second tweet by proclaiming that:

"It's time for Amjad to go by a single name like Elon and Beyonce."

Replit had recently reported hitting $400m ARR which, if true, means it has overtaken its self-proclaimed rival Lovable, which seems to have fallen behind in recent months as it is now seen more as a place to build prototypes than full-stack apps.

GPT 5.4 is out!

OpenAI decided to stealthily drop GPT-5.4 at midnight.

GPT-5.4 excels at three things when compared to other state-of-the-art models, including other OpenAI models:

1. Computer use

GPT-5.4 has been given major upgrades in computer use, scoring an impressive 75% in OS-World. GPT-5.2 scored 47.3%, and after using it for a few cron-job tasks, I can definitely say GPT-5.4 is now my go-to model for computer usage.

2. Tool usage

OpenAI has implemented a new feature called "Tool Search." This feature involves giving the model a predefined list of tools that it instantly has access to if needed. This reduces latency because the model does not have to spend time searching for a tool, and it also reduces cost because fewer tokens are spent on that search step.

3. Steerability

If there was one standout feature for GPT-5.4, it would definitely be its new steerability feature. It is available for the thinking versions of GPT-5.4, where a user can inject a prompt midway through the agent's thinking, which the agent then incorporates into its context and answer.

So instead of stopping the agent, resubmitting the prompt, or waiting until the run is done before asking the thing you left out, you can now quickly and seamlessly add it into the agent chain.

The timeline

Thanks for stopping by!

Prefer the hosted version? Read the original post on Beehiiv.