The Loom Method: How to Find Your First AI Product in 43 Minutes

Most business owners looking to build an AI product make the same mistake: they start by asking "what should I automate?" That question sounds reasonable. It produces terrible answers.

You get a list of vague candidates — "maybe we automate our onboarding" or "we could do something with customer emails" — none of which ever become real products because they're too broad, too undefined, and too easy to deprioritise when something urgent comes up.

Greg Isenberg, one of the more careful thinkers on AI product discovery, has a different method. He doesn't ask what to automate. He watches how someone works — specifically, a Loom recording of them doing the job — and looks for the moment where the mouse repeats for more than 30 seconds. That repetition, he argues, is the product. Not an idea for a product. The product itself, hiding in plain sight.

The title of this post comes from a YouTube video where Gumloop's CEO built a working AI agent in 43 minutes, live on screen. But the more important idea is what you do before you build anything: identifying the right workflow to automate in the first place. That's where most founders fail, and where the Loom method changes the game.

Why "What Should I Automate?" Is the Wrong Question

The problem with open-ended brainstorming about AI products is that it produces ideas based on what sounds plausible, not what is actually painful and repeated. Founders end up building solutions to theoretical problems they've never actually observed in the wild.

The Loom method sidesteps this by using evidence. You're not hypothesising about what's annoying. You're watching what's actually happening, in real time, on someone's screen.

Isenberg's specific instruction: open someone's "how-to" Loom — the kind anyone records when training a new hire or explaining a process — and watch for any section where the cursor repeats itself for more than 30 seconds. That repetition is what he calls the extraction point. It's the task that's being done the same way, over and over, that doesn't require judgment, only execution.

Where the cursor repeats, there's usually one of three things happening: the person is copying data between tools, reformatting information into a different structure, or running the same lookup or check on a rotating list of inputs. All three are automatable. All three are frustrating enough that someone will pay to have them solved.

The insight is not that repetitive tasks are automatable — that's obvious. The insight is that you can identify them reliably in under an hour if you know what to look for. The Loom method is a repeatable process for finding product ideas, not just an abstract principle.

The Five Most Common Extraction Points

After applying this framework across dozens of processes, the same categories surface repeatedly. If you're watching someone work and looking for extraction points, these are the five workflows where the mouse repeats most often:

Lead qualification and research. Someone opens LinkedIn or a CRM, reads through a profile, checks a few signals, then copies notes into a spreadsheet or updates a status field. The same 12-step process, repeated for every new prospect. This is one of the most commonly productized automations — and one of the most valuable, because it directly touches revenue.

Content repurposing. A recording, transcript, or article gets manually reformatted into a newsletter, LinkedIn post, or email sequence. The same material gets restructured through the same template every time. Pure execution, no judgment required.

Internal reporting. Someone pulls numbers from two or three tools, pastes them into a template, writes a brief narrative, and sends it to a manager or client. Weekly or monthly, same structure every time.

Customer onboarding steps. New clients go through a sequence of manual touchpoints — welcome emails, setup instructions, check-in calls, follow-ups — that someone coordinates by hand. Often tracked in spreadsheets long after the business could afford a better system.

Proposal and document generation. A sales or account rep pulls information from a CRM or brief, writes a customised proposal using a mental template they've memorised, then formats it into a document. Looks creative. Mostly isn't.

None of these are surprising. What's useful is having a clear list before you sit down with the Loom — so you recognise the extraction point when you see it instead of watching it go by.

What Makes It a Product, Not Just a Script

Finding an extraction point is the first step. The next question is whether the automation you build from it is a one-off script or a real product with recurring revenue. Most builders stop at the script. The ones building $50M companies like Gumloop are doing something different.

The difference is context accumulation. A script automates the process once. A product gets smarter every time it runs because it's absorbing proprietary context — the client's vocabulary, their edge cases, their preferences, their internal abbreviations. That accumulated context is the moat. It's why a customer who's been using your product for six months can't easily switch to a competitor's tool that runs the same underlying automation.

Isenberg's productization logic follows this pattern directly: capture the workflow, then load it with the customer's proprietary context — their FAQ, their tribal acronyms, their edge-case examples — formatted into plain text or JSON that a general LLM will never have seen. The product isn't the automation. It's the automation plus the context asset it builds over time.

A practical test for whether your extraction point can become a product rather than a script: ask whether the output gets meaningfully better as the system sees more of that specific customer's data. If yes, you have the foundation for something sticky. If not, you have a commodity tool that a competitor can replicate in a weekend.

Companies like Zeda.io demonstrate this pattern in product intelligence — they turned the repeatable process of organising customer feedback into a system that compounds insight over time. The extraction point was obvious (product teams manually triaging feedback). The moat was the context the system accumulated about each company's product vocabulary and priorities.

The Validation Step Before You Build

Before writing a line of code or spinning up an agent, there's a validation step that most founders skip — and it's the one that separates the $13k/month AI automation business from the project that never found a customer.

The validation test is simple: find three people who currently do the workflow you want to automate, and ask them this question — "if I could take this entire process off your plate reliably, what would you pay per month?" Don't describe your solution. Describe the outcome. The answer tells you two things: whether the pain is real, and whether it's specific enough that buyers can immediately picture the before-and-after.

If they hesitate or give you a number under $100/month, one of two things is true: either the workflow isn't painful enough to monetize, or you're describing it too abstractly. Sharpen the description and ask again.

A strong extraction point has these characteristics: it repeats weekly or daily, has a clear before/after outcome, uses bounded human judgment (enough to be non-trivial, not enough to require genuine expertise every time), and touches money, speed, or compliance. The case study of the founder charging $13k/month with a single AI automation hits all four — the workflow was specific, the value was obvious, and the buyer didn't have to stretch to understand what they were getting.

The AI agents market is projected to reach $48.3 billion by 2030, growing at 43% annually. The question is not whether demand exists. It's whether you're targeting a workflow that's specific enough to produce a clear value proposition — rather than a vague "we use AI to make your operations faster" pitch that sounds compelling and converts nobody.

The Action This Week

Find one Loom — yours, a teammate's, or someone in your industry who's posted a "how I do X" walkthrough. Watch it from start to finish. Note every moment where the cursor repeats for 30 or more seconds.

If you find one, write a one-sentence description of the automation: "An agent that takes [input] and produces [output] for [who] every [frequency]." If that sentence is specific enough that you could quote a price for it, you have an extraction point worth validating.

Then ask three people who currently do that job whether they'd pay to have it handled. The market exists. The only question is whether you pick the right extraction point to enter it.

Summary

Most founders waste months searching for the right AI product idea because they start from a blank page. Greg Isenberg's Loom method replaces that open-ended search with a concrete, repeatable technique: watch a workflow recording, find where the cursor repeats for 30+ seconds, and treat that repetition as your extraction point. The five most common extraction points — lead research, content repurposing, internal reporting, customer onboarding, and document generation — account for the majority of profitable AI agent products being built right now. What separates a product from a script is context accumulation: the agent that gets smarter from each customer's proprietary data builds a moat that's hard to replicate. The validation step is simple: describe the outcome to three potential buyers and ask what they'd pay. If the answer is clear and specific, you have a real product opportunity hiding inside a 43-minute Loom.

FAQ

Q: Do I need technical skills to build an AI agent product from an extraction point?

A: Not for the discovery phase, and increasingly not for the build phase either. Tools like Gumloop, n8n, Make.com, and Flowise let non-technical builders turn identified workflows into working automations without writing code. The extraction point method is entirely non-technical — it's about observation, not engineering. Once you've validated demand, you can decide whether to build yourself or partner with someone who can handle the technical implementation.

Q: How do I find Looms to watch if I don't have a team doing the workflow internally?

A: YouTube is the most accessible source. Search for "[industry or role] + how I do [task] + walkthrough" and you'll find dozens of real practitioners demonstrating their workflows on screen. SOP libraries, Loom channels, and onboarding videos shared in industry communities (Slack groups, subreddits, Facebook groups) are also rich sources. The goal is to watch real practitioners doing real work — not polished explainer videos, but the messier "here's my actual process" style recordings.

Q: What if the extraction point I find is something a lot of other people are already automating?

A: Competition in the category is actually a positive signal — it confirms the pain is real and the willingness to pay exists. The question is whether you can serve a specific vertical or customer type better than the general solution. Specialisation almost always beats generalisation in AI automation: an agent built specifically for real estate lead qualification will outperform a generic research assistant for a real estate team, because it accumulates context about their specific market vocabulary, their qualification criteria, and their edge cases. Find your extraction point, then niche down on who you're serving.

Sources

• Greg Isenberg on LinkedIn — "Productizing AI agents is the biggest zero-to-one play"
• Greg Isenberg — Multiply Your Output With AI Workflows
• YouTube — I built an AI Agent in 43 min to automate my workflows (Gumloop CEO)
• The Top Voices — Gumloop Raises $50M Series B
• YouTube — How She Built a $13k/m Agency Using Just ONE AI Automation
• TechCrunch — After selling his last startup to Google, this founder now wants to automate mundane tasks with Relay
• BCC Research — AI Agents Market to Grow 43.3% Annually Through 2030
• SaaSLogic — How to Price an AI Agent
• Towards Data Science — How I Built Business-Automating Workflows with AI Agents
• GREG ISENBERG on X: Productizing AI agents is the biggest zero-to-one play