The Cursor Blinks For Thee
How the future feels in Silicon Valley right now
This is what it feels like to experience the future in Silicon Valley right now.
A future that all of humanity is going to experience shortly.
Your profession — the one you prided yourself on, the highest-compensated profession in human history with eight of the ten richest humans, and forty of the top hundred, deriving their fortunes from it, all within the last 30 years, — is now being performed by this blinking cursor.
You are a little dazed.
It does things that used to require years of experience and admission to the best engineering schools in the world. My high school in Singapore produced one International Math Olympiad winner in the four years I was there. There have been roughly 1,800 IMO gold medalists since 1959. At least fifty are in the Bay Area. Many work in AI.
And the cursor is matching them.
At first you test it. You probe the borders of your own expertise. You ask it things you already know, then things you half-know, then things you could find out for yourself if you bothered to ask Google and wade through 30 search engine optimized pages with bits of information here and there that had to be collated into a comprehensive answer. Eventually you ask it to do something just outside your field.
That is where the trouble begins.
Because once it works there, you start building tools.
It creeps up on you. At first it is a monitor for your personal Robinhood trades. Then it is something to clip YouTube videos, something you have always found annoying to do quickly.
Then it is some other small tool you once wanted, but never wanted enough to build or hire for. Before, that would have meant another $10–$100 SaaS subscription. Another tool to learn. Another interface that almost, but not quite, did what you wanted. Another charge quietly ringing up your credit card after you forgot you had subscribed.
Then comes the relinquishment.
At some point, you are too busy to manage everything, so you drop a large task on the cursor and leave it alone.
This is where the cursor stops being a tool and starts becoming a presence.
Because the cursor does it.
It goes through your emails and writes thank-you notes to all the people who came to your kid’s birthday party. It helps you find that subscription you signed up for and have been trying to cancel for months. It cleans the little corners of your life you had quietly given up on.
And you are hooked.
This is where many CEOs are now.
It is quite, quite amusing to me that one of the companies in this revolution is called Cursor.
Because now the cursor blinks at you every day.
The first thing a good number of us are doing is organizing our thoughts.
The process starts with the realization that there is some part of your mind you always wanted to externalize. Some complete memory of a certain thing. In Andrej Karpathy’s case, it is a map of every concept in every research paper he has ever read, or should read, built on the conviction that insight comes from connecting scattered pieces of the literature into new meaning.
In Garry Tan’s case, it is a full memory of every person he has met, as he meets thousands of people a year. VCs call this the “personal CRM,” and it is something that has been hunted for decades, like a holy grail.
So a good number of people are busy constructing these stores for themselves. There are debates on how to organize them. For now, the answer seems to be simple text files the cursors can read.
At about this point, even the cautious among us begin to give in.
We release the cursors into our workspace.
We have no patience to be bothered every ten minutes over whether we would like to approve something as insignificant as file deletion. This is called going “full-auto.”
Going full-auto is when you unleash the cursor. You are no longer sitting there, watching it work. You are allowing it to significantly alter the core product of your profession.
And like a father watching his child learn to cycle for the first time, you take a deep breath and let go.
And the cursor is doing it.
It is handling it.
When you come back, it has erected a reasonable addition to your body of work.
Your awe is tempered by the imperfections still evident in the work. Some are matters of opinion rather than accepted practice. Some are real flaws. But on the whole, with a little doctoring, it passes.
This is where we are as coders right now.
So now we find ourselves wondering what else we can build.
If it is good enough to provide top-quality work in my field, is it good enough to provide top-quality work in the field of the person selling to me? Or the field adjacent to mine? Or the field necessary for mine?
For example, medical researchers are often not trained as professional statisticians. In large institutions, they hire one. In smaller labs, they make do with software, templates, and internet advice.
Many a million-dollar study brings on a statistician as an afterthought, only to be told that the research they had done is useless. The study design cannot answer the question they asked, in effect they asked a different question from what they mean to ask. The wrong things were measured. The data, painstakingly collected, cannot be made to confess what it never observed.
But now every lab can have a competent statistical collaborator on demand. The first step to designing the study is a long chat with a team of cursors: a brainstorm organizer, a research assistant, a statistician, a reviewer. The gap between the small lab and the large one narrows.
And not just for the math piece. The cursor can be asked to help from the idea to the final product, across the whole chain of work.
You usually require a nudge at this point to hand off more work to the cursor.
For me, this happened on the 5th of May, when Sam Altman gifted those turned away from the GPT-5.5 launch party with a 10x increase in usage limits.
For the first week I was skeptical.
What would one even use such ludicrous amounts of tokens for?
I didn’t have that many tasks I could send the cursor to.
Or did I?
Because if one were careless with one’s usage, and did not mind missed turns, detours, unfamiliar terrain, and rejecting bad work without agony over sunk cost, one could perhaps journey to a place one had always wanted to go, but never had the time or energy to reach.
You embark on a Project.
A Project is something you have wanted to build for a while. It has been gnawing at you. It would have involved far too much work for the benefit it might deliver. Or it was risky, with a low chance of success, and you discarded it.
But now, with Sam’s Gift, you can throw the cursor at it and see what it can do.
It takes several tries.
You set it off with just an idea. Then you get the results and guide it on what to do next. But the initial kernel expands into long to-do lists, and the cursor starts to get confused. Sometimes it solves something once, then solves it again later in a different way.
It is as though you are the architect, and the cursor is your first workman.
First you sketch a picture of what you want. Then it builds it. And it is okay. But you want another window over there. When the cursor fixes that, the floor collapses here.
Then one cursor becomes many.
You have forty chats open. One is watering the flowers. One is taking out the trash. One is adding a west wing to the house. Another has forgotten the original blueprint and is quietly building a second house next door.
So the old tools of software civilization return: git, GitHub, linting, formatting, tests, pull requests, rules.
We built these systems to coordinate humans.
Now we are using them to coordinate cursors.
The job changes again. For the first time in decades, you stop writing code. The things you write now are prompts, they are design specs, they are reviews once the work is done. You are now the owner providing the keys to castle, the vision of what must be built, the critical reviewer accepting the product. You approve the architect’s plans, you monitor speed and cost, you wait to see if the cursors get stuck.
Now you organize a team.
This is where Garry Tan was about a month ago.
You build out the structure. Let there be a single to-do list. Let one cursor take the list and describe, in detail, what needs to be done. More importantly, let it define what kind of result is acceptable. Let another cursor grab the task and hand it to another cursor whose job is to do the thing.
You give the doer its rules. You tell it what it is allowed to do. When it completes the task, it must provide a report: what it changed, what failed, what pitfalls it encountered, and what should be stored for the next time it faces the same issue.
It is allowed to come back to you when it needs something: keys, access, a decision, a judgment call.
And then you send the team off.
And wonder of wonders, it works.
Overnight, they build something that would have taken you weeks.
In normal corporate bureaucracy time, it would have taken months just to pitch the idea and get the resources. It would have taken a team of three or four people weeks to build what you now have in the morning.
You are a bit dazed.
This is roughly where people at Anthropic and OpenAI have been for a couple of months now.
This is where Elad Gil says, “We are likely in very early lift off & exponential.”
For now, it is the most ambitious who are taking advantage of this.
SemiAnalysis, a small firm of semiconductor analysts, now tracks the entire industry in the US and Asia with a team of fewer than twenty, while the research departments at the Wall Street banks are behind.
Ramp began as a company for managing corporate financial plumbing. It is now beginning to look like the financial nervous system of new corporate entities.
But the tokens are expensive.
Top people are spending tens of thousands per month at this point.
Token leaderboards inside firms, ridiculous and gameable as they sound, have forced people much like me to explore the use of cursors more widely and with more ambition.
And then the trial ended.
Sam’s Gift came to an end last week, the 5th of June, exactly one month later.
I had been watching the date approach with anxiety. I know myself. It is going to be very hard not to tell the cursor to do more things.
I have been wryly amused observing myself fall prey to the addiction.
Not a gift.
A free first dose of heroin, selectively delivered to the top 8,000 or so superfans of the company, the people who had volunteered with enthusiasm to attend a party for the GPT-5.5 launch.
What a perfect audience.
A test case of the most susceptible. The most open to the message.
Enthralling.
But wait. Just one more thing.
Yesterday, Anthropic announced that the cursors are helping to build better cursors.
The cursors are building the factories that will build better cursors. Some of the parts coming off that line are already better than the ones made by the best humans.
Which means they have become capable enough to increase the capability of the next generation.
This is the dream that floats above Silicon Valley today.
Coders cycle to work with their laptops running in their backpacks, slightly ajar, keeping their cursors alive.
CTOs, normally trapped between infinite founder vision and finite engineering budgets, are discovering a new constraint. Not headcount. Not money. Definition and review. Can they specify what they want clearly enough? Can they judge what comes back quickly enough?
Now I am become bottleneck, destroyer of velocity.
The chief technical unblocker is now the technical block. The cork in a shaken up champagne bottle on the verge of productive explosion.
Some sleep less now. Some sleep in two-hour blocks, waking to check whether their cursors are stuck.
It is a dream of unleashed technical ambition. It is a rising confidence that whatever can be built will be built with a subtle undercurrent of fear that some of those things shouldn’t.
And all the while, we watch the cursor on our screens.
Blinking.
Beckoning.
