Someone asked GPT-6 to draw a portrait in Google Calendar.

Another post asked Astra to open Paint in a browser and draw the user.

Both ended up in my bookmarks. A calendar is a ridiculous place to request a portrait, which is probably why that example is so memorable. The Paint post makes the connection to operating an interface more direct.

The posts alone do not establish exactly how every action was performed. But they point toward a useful question: how much can an AI do inside software we already have?

I am Alex. Welcome to ShortCu8 by Innov8.

Lets Dive Deep🐰🐰.

An application does not necessarily need its own AI feature for an agent to use it.

If the agent can access the interface, understand its controls and check the result of its actions, it has a route to doing work there.

The basic loop is:

Observe -> act -> check -> adjust -> continue.

A useful computer agent needs to keep that loop working across the whole task, including the parts that do not go as expected.

Imagine asking an agent to update a presentation using numbers from a spreadsheet.

It has to locate the right data, understand which slides need changes, preserve the layout and check that the finished deck still makes sense.

Then a dialog appears. Or a table does not fit. Or the spreadsheet contains two versions of the same number.

Completing the job requires the agent to stay oriented through those interruptions. It needs to recognise a mistake, recover from it and decide when it needs your input.

This is why speed and reliability matter together. An assistant that needs rescuing after every few actions still leaves you supervising the entire process.

OpenAI's Astra announcement describes improvements in computer use and work inside documents, spreadsheets and specialised software.

In its OSWorld 2.0 latency simulations, OpenAI reports 72.6% performance at roughly 40 minutes per task, compared with 65.7% at roughly 75 minutes for GPT-5.6 Sol. That is about 47% less time in the reported comparison.

These are vendor-reported evaluation results. They do not mean Astra completes 72.6% of whatever we ask it to do on our own computers.

But they support a more useful claim than a single impressive clip: OpenAI reports improvement in both task performance and the time needed to achieve it.

The surrounding software matters too. OpenAI says it has also updated the Codex system that handles computer use. Some gains therefore come from the combination of the model and the tools around it.

Consider an old supplier portal that requires copying information from invoices into several forms. Or a reporting tool where someone has to repeat the same export procedure every week.

An agent capable of operating those interfaces could help without waiting for each vendor to build a dedicated AI integration.

For creative work, the useful result could be an editable file inside the tool you already use: a presentation with its original layout preserved, or a scene you can continue adjusting.

Your bookmarks also contain Ilker's Blender example, which he says he assembled with Astra and PATINA. That belongs to the wider story of AI working with professional tools, although the post does not establish that it used only visual controls.

Agents can write code, call an application's programming interface or interact with its visible controls. Those routes can work together. A finished output does not tell us which route produced it.

The benefit is that more of the work can happen where the result actually needs to live, instead of ending as instructions in a chat window.

A short recording usually shows one selected attempt. It may leave out waiting, failed runs or help from the person making the demo.

For everyday use, the test is whether the agent completes the job correctly with an acceptable amount of supervision.

Does it notice that it edited the wrong slide? Can it recover when a page changes? Does it stop at the point where a decision belongs to you?

Those details determine whether computer use saves effort. Producing more actions is not enough if someone has to inspect and repair each one afterwards.

We read every reply - just reply to this email and let us know how we can improve !

#AD1

For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.

LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.

The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.

Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more.

#AD2

Every sales leader knows the feeling. You walk into a pipeline review with a number you believe in, and twenty minutes later, you're defending every line item to a CEO who just wants to know what's actually going to close.

HubSpot Sales Hub ends that conversation. Every deal, every rep's activity, and every buyer signal are all in one place and updated automatically. So your forecast is built on what's actually happening. And when you present that number, you can stand behind it.

Get a demo