Big Idea

The argument in one line.

A screen-watching agent that logs activity every five seconds and waits for an inefficiency to repeat three or four times before suggesting a fix turns invisible, sub-minute habits into a compounding, provable time-and-money return.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use a computer for most of your workday and suspect small repeated actions are quietly costing you time.
You manage a team and want a low-effort way to spot inefficiencies in new hires' workflows without shadowing them.
You're comfortable running a background script or agent and want a concrete example of an AI agent with a measurable, personal ROI.

SKIP IF…

You're looking for a polished, ready-to-install product — this is a personal script Nick built for himself, not a packaged tool.
Your work happens mostly off-screen (physical labor, meetings, phone calls) where screen capture can't see the bottleneck.

TL;DR

The full version, fast.

Nick Saraev built a small AI agent that screenshots his screen every five seconds, logs roughly 8,640 lines of metadata a day, and runs a cheap model over that log to spot repeated low-value actions. Once a behavior shows up three or four times, the agent proposes a concrete upgrade — a hotkey, a Chrome extension, a cheaper tool — and he can approve it in seconds. In under an hour of use it caught a redundant checking pattern, a five-minute daily task worth automating into a hotkey, and a voice tool with five times the latency of a free alternative. The system took under five minutes to set up and has already returned roughly thirty minutes of his day, with an estimated ceiling of two to three hours once he acts on everything it has flagged.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:10

01 · A 5-minute system that buys back 30 min/day

Nick introduces the system: an AI agent watches his screen and proposes workflow fixes; setup took under five minutes and already saved roughly half an hour a day, with an estimated ceiling of 2-3 hours.

00:10 – 00:56

02 · How it works: a screenshot every 5 seconds

The agent screenshots the screen every 5 seconds, logs to a daily folder database, runs a cheap model to summarize and count repeat low-value behaviors, builds an observations ledger, then Nick periodically asks it how to economize his workflow.

00:56 – 01:41

03 · 3 real upgrades from the first hour

Three concrete examples surfaced in under an hour: a morning digest replacing a repeated polling loop, a hotkey for idea capture wired to Raycast and Linear, and a suggestion to combine views in his project manager.

01:41 – 02:23

04 · The hidden GDP buried in tiny inefficiencies

Nick argues that seconds-long frictions — mousing to a menu instead of using a hotkey — compound across millions of workers into real losses in economic output.

02:23 – 03:03

05 · The voice tool swap that paid for itself

The agent flagged that his voice-transcription tool 'Aqua' had ~250ms latency versus ~50ms for a free local alternative ('Hex'); switching saved both money and daily micro-delay at scale.

03:03 – 03:44

06 · New hires + the Chrome extension it wrote for him

Applied to onboarding, the same system can watch a new hire's first day and recommend a tailored fix — Nick describes it writing him a Chrome extension after spotting a repeated tab-opening routine.

03:44 – 04:34

07 · Beyond the desk: watching your whole life

Nick floats extending the same logging-and-ledger loop to physical life — a camera capturing a room over 3-7 days to catch posture, eye strain, or desk-ergonomics habits.

04:34 – 05:11

08 · Under the hood: metadata, frames, and the ledger

Technical breakdown: ~8,640 lines of text metadata a day (cheap), ~20 sampled frames for vision-only cases, an observations ledger that confirms a pattern after 3-4 repeats, then a suggested upgrade.

05:11 – 05:45

09 · Run it free on your session limits

Nick suggests running the analysis loop once every 24 hours, including overnight, to use otherwise-idle AI session capacity.

05:45 – 06:22

10 · Get everything below (scripts + diagrams)

Nick offers the script, the background screen-watcher, and the diagrams in the video description for viewers to set up the same system themselves.

Atomic Insights

Lines worth screenshotting.

A screenshot-every-five-seconds logging agent found three concrete workflow fixes inside the first hour of use.
The system waits until an inefficiency has repeated three or four times before proposing a fix, filtering out one-off noise.
8,640 screenshots' worth of metadata a day is cheap because it's processed as text, not images — vision analysis is reserved for roughly 20 sampled frames where text can't capture what happened.
Swapping a paid voice-transcription tool for a free local one cut per-use latency from roughly 250 milliseconds to 50 milliseconds and eliminated a recurring subscription fee.
At an estimated 500 uses a day, saving 200 milliseconds per use recovers a few real minutes daily — proof that sub-second frictions compound.
A five-to-ten-minute manual idea-capture routine became a single hotkey wired through Raycast and the Linear API.
The same agent pointed at a new hire can, after roughly a day of activity, recommend a Chrome extension or shortcut tailored to their specific repeated tasks.
Small per-action inefficiencies — a few extra clicks, a menu dig instead of a hotkey — scale from personal annoyance to measurable lost output once multiplied across an organization's headcount.
The agent's logic could extend past the desktop: recording a room over three to seven days and feeding those frames into the same pipeline could surface physical habits like poor posture or eye-level monitor placement.
Running the analysis loop before bed uses idle AI-session capacity that would otherwise go to waste overnight.

Takeaway

A screenshot every five seconds can find thirty minutes a day

WHAT TO LEARN

Waiting for an inefficiency to repeat three or four times before acting turns invisible, sub-minute habits into a provable, compounding time-and-money return.

02How it works: a screenshot every 5 seconds

The agent's only sensing action is a screenshot every five seconds — everything else runs on that raw capture, so the setup cost is a single background script.
A cheap model does the first pass over the log to summarize activity and count repeated low-value actions, reserving expensive reasoning for only the patterns worth flagging.
The system only surfaces a suggestion after building a durable ledger of observations — it doesn't react to a single occurrence, it waits for a repeated pattern.

033 real upgrades from the first hour

A repeated pattern of checking multiple web pages throughout the day got replaced with a single batched morning digest — same information, one interruption instead of eight.
A five-to-ten-minute manual idea-capture habit became a one-hotkey voice-to-Linear pipeline, cutting a daily task down to a single keystroke.
Even a project-manager view got flagged: the agent suggested combining several separate views into one simpler workflow.

04The hidden GDP buried in tiny inefficiencies

A single avoidable habit — mousing to a corner menu instead of using an existing hotkey — can cost roughly fifteen seconds for no reason, and that's the whole argument: small enough to ignore, repeated enough to matter.
The claim scales from personal to macro: multiply a fifteen-second habit by hundreds of thousands of workers and the losses become large enough to matter at a national productivity level.

05The voice tool swap that paid for itself

Swapping a paid voice-transcription tool for a free local alternative cut per-use latency from roughly 250ms to 50ms while eliminating a recurring subscription fee.
At an estimated 500 uses a day, a 200ms-per-use savings compounds into a real, measurable chunk of reclaimed time — not just a monthly dollar figure.
The video states the swapped tool's monthly cost two different ways ($15 and $30) — a reminder to verify exact numbers before repeating a specific claim like this one.

06New hires + the Chrome extension it wrote for him

The same logging loop works on someone else's workflow: pointed at a new hire, it can flag a recurring pattern within about a day of use.
When the agent spotted a recurring set of tabs being opened manually, it wrote a Chrome extension that opens them all at once — the fix, not just the diagnosis.
The 'small saving in isolation, massive saving in aggregate' framing applied to GDP earlier gets reapplied here to onboarding a single employee.

07Beyond the desk: watching your whole life

The same log-then-ledger structure could extend past the screen — a camera recording a physical space over three to seven days feeding the same pattern-detection loop.
Candidate physical patterns include posture while working, squinting from poor lighting, or a monitor that isn't at eye level — things a person is unlikely to notice about themselves.

08Under the hood: metadata, frames, and the ledger

Roughly 8,640 lines of metadata get logged per day, and because it's text rather than images, the ongoing cost is close to free.
Vision analysis is reserved for a much smaller sample — around 20 frames a day — used only where text metadata can't capture what happened.
A pattern only gets confirmed, and only then suggested as a fix, after appearing three or four separate times in the ledger.

09Run it free on your session limits

Running the analysis loop once every 24 hours is enough cadence to catch and act on new patterns without constant manual review.
Running the loop overnight, before bed, uses AI session capacity that would otherwise sit unused until the next billing window resets.

Glossary

Terms worth knowing.

Observations ledger: A running, durable log of flagged inefficiencies that the agent updates each pass; a behavior only becomes a confirmed pattern once it appears three or four times.
Cheap pass: An inexpensive AI model run over a day's screenshot metadata to summarize activity and count repeated low-value actions, before anything reaches a more expensive analysis step.
Session limit window: The usage allowance for an AI subscription tier over a rolling period; running background analysis overnight uses capacity that would otherwise sit idle.

Resources

Things they pointed at.

00:00toolFable

02:23toolAqua (voice transcription tool)

02:23toolHex (free local voice-transcription alternative)

01:17toolRaycast

01:17toolLinear

Quotables

Lines you could clip.

00:45

“I have gotten probably at least half an hour back in my day in just the last few days alone.”

credible-skeptic setup ('optimized knowledge worker') before the payoff claim→ TikTok hook↗ Tweet quote

02:25

“There's a massive amount of GDP or gross domestic product buried in minor inefficiencies like this.”

punchy, quotable macro claim→ IG reel cold open↗ Tweet quote

05:40

“Your session limit... is basically a scarcer resource than gold at this point.”

sharp one-liner about AI usage economics→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

analogy

00:00So since we all have Fable for the next few days and a lot of people are probably wondering how to use it, here's an extremely simple system that anybody in any organization can apply that'll significantly improve their productivity using a computer. To make a long story short, it's a system where you work, Fable or another AI watches your screen and captures a screenshot every five seconds.

00:18You log that to a daily folder database, pass this to a really cheap model to summarize and then count any repeat behaviors that are low value. You make a ledger of notes and observations, and then you just ask Fable repeatedly once a day or once a week, whatever cadence you wanna spend your API tokens on.

00:34How can I economize my workflow? It sounds really simple, and it took me less than five minutes to implement this. And it works really well.

00:41I have gotten probably at least half an hour back in my day in just the last few days alone. I realistically think I'm probably going to get closer to two to three hours of my day back. And you gotta keep in mind that, like, I consider myself a pretty optimized knowledge worker.

00:54I've been working with computers for basically my whole life. I mean, here's an example of a quick little brief, and these are three ideas that this recorded in less than an hour of actual screen time. Morning digest instead of a polling loop.

01:04It found that I was accessing web pages in a relatively cyclical manner, checking for various updates. And it suggested, why don't I create you a high quality morning digest instead that includes all the information that you'd get, but now you batch it in one instead of eight times. The next is a hotkey idea capture.

01:19Right now, I have sort of a loop where I'll design content ideas throughout the day. It took a system that was previously taking me five to ten minutes additional per day, and it turned that into a Raycast script command with a linear API pass through. So basically, I just press one hotkey, dictate a line, and then immediately get back to work.

01:35It gave me ideas surrounding a cross team view in my project manager, basically just telling me, hey, you should combine all these into a much simpler workflow. And the reason why I wanted to make this video is because I legitimately think there's a massive amount of GDP or gross domestic product buried in minor inefficiencies like this.

01:50Again, the fact that you have to mouse your thing to the top left hand corner of the screen, click two or three times, consume fifteen seconds for no discernible reason at all when a hotkey exists. These small inefficiencies compound over time across hundreds of thousands and millions of people in organizations to legitimately reduce the gross domestic product of your country.

02:08If we can get as many people on work optimizations as humanly possible, we could basically shave off all of the current excess usage or excess wastage of these actions and funnel all our time and energy towards things that actually move the needle for businesses, for people, and for government. I showed those examples to you because I want you to see realistic gains that you will get in the first few hours of using it.

02:28But I've also gotten gains that are directly translated to financial return. I mean, was previously using a voice transcript software called Aqua that was costing me a fair amount every single month. It was also taking around two hundred and fifty milliseconds every single time that I used it in order to populate text.

02:43So not only did it save me $15 a month, this solution, which is previously taking me about a quarter of a second every time I voice transcribed, which it picked up that I voice transcribed hundreds, if not thousands of times per day, took me down to fifty milliseconds, saving me something like two hundred milliseconds times 500, winning me back a few minutes every day, plus $15 a month.

03:03And I know all of this stuff sounds stupid, but imagine a strategy like this applied to new people in an organization. They come on board, they don't know any of the SOPs, and then immediately, you know, after a day or so of doing something, Fable analyzes the way that they do their workflow and recommends them a Chrome extension, which saves them five minutes a day.

03:21That is a small saving in isolation, becomes a massive saving in aggregate. And I know this because that's literally what happened here. I was doing the same basic procedural workflow every single day where I was opening up a certain number of pages constantly.

03:33Fable just said, why don't you just open up all of those pages simultaneously with a Chrome extension? I've just written one for you. Here you go.

03:39None of these things seem really big, but they stack up. And now I'm saving a fair amount of time every day. So I personally think this is probably going to be the future of all sort of work optimization, least digitally.

03:50And I think you can actually apply this more broadly to your life. Like, I'm actually considering, I know this sounds wild. Again, installing a camera in the top right hand corner of my room, and then just having it take screenshots every five or ten seconds.

04:01Run that for a three to maybe seven day series, and then feed that into Fable alongside the sort of annotation workflow. Can you imagine just how many minor behaviors I'm probably doing?

04:11Hell, it could be me leaning over to scratch my back or something like that. Fable would pick up on that and be like, hey, is there something going on over there? What's going on?

04:18How can I help you? Maybe it's the way that I'm leaning in my chair leading to back pain. Maybe I'm constantly squinting because the light is in my face.

04:26Or maybe, you know, my monitor is not necessarily at eye level. I'm kind of off to one side. All of these things seem really small in isolation.

04:32Again, they become really big in aggregate. So the actual system is gonna consume something like 8,640 lines of metadata.

04:40It's text, so it's basically free. You'll sample around 20 frames, vision only where text can't actually tell things. Again, you'll modify an observations ledger, which is just a big list of observations.

04:51And then if it's seen more than three or four times, it'll confirm the pattern of poor behavior or pattern of inefficiency, and then suggest an upgrade. So, you know, hey, use the hotkey.

05:00You mouse through the menu four times today. Hey, this $30 a month voice tool, here's a free local one. Hey, this five to minute this five minute task repeats three to four times a day.

05:08Do you want a Chrome extension that does it for you? The really cool thing about this workflow as well is I know everybody here has a certain amount of Fable session limit available. Um, you can run this sort of analysis in the background for yourself really easily.

05:19Like for instance, once you populate this folder of your screenshots, which is occurring constantly in the background every few seconds, like it's taking screenshots of me and this right now, you could just run a loop once every twenty four hours where Fable like wakes up on your computer, quickly analyzes everything, and then lets you know, hey, you know, I think you should do this.

05:36I think you should do this. I think you should do this. You could also run this before you go to bed to utilize your session limit window, which as we know is basically a scarcer resource than gold at this point.

05:45Lots of ideas that I just wanted to throw at you, but if you guys wanna get this script and everything else, I've actually just included all this down below. It includes a script that you can feed into Fable to have it build it for you. I'm also gonna include the ongoing screen watcher script that basically just runs every five seconds in the background, that that'll obviously require some configuration if you're on a Mac or if you're on a Windows because of different accessibility permissions.

06:05I'm also gonna include just these diagrams and these files below so you guys have everything you need, and more or less whatever else I can just to give you guys that value. So not a super crazy idea, I would hope. I want you guys to play around with this idea and let me know your thoughts down below in the comments.

06:17Thank you very much for watching. Have a lovely rest of day, and I'll catch all y'all on the next video. Cheers.

The Hook

The bait, then the rug-pull.

Nick Saraev spent five minutes wiring together a screen-watching AI agent — by the time he finished recording this video, it had already found him thirty minutes a day in reclaimed time.

Frameworks

Named ideas worth stealing.

00:21model

Screenshot -> ledger -> suggestion loop

watches your screen
capture script every 5 seconds
daily folder db (frames + log.json)
cheap pass: summarize + count repeat behaviors
notes + observations ledger (durable)
ask the agent: how can I economize my workflow?
suggests: hotkeys, faster tools, or a custom Chrome extension

The full loop Nick sketches on the whiteboard: continuous low-cost logging feeds a cheap summarization pass, which accumulates into a durable ledger, which only surfaces a suggestion once a pattern is confirmed.

Steal forany personal or team workflow-audit system — the loop generalizes past this specific tool

CTA Breakdown