# Where are the agent founders?

Source: https://x.com/huntermmonk/status/2116185515837784543
Kiwi News: https://news.kiwistand.com/stories/Where-are-the-agent-founders?index=0x6aec54273c546f320da82096a0b00268db34f72306f40a45ef3ff412d2365ed3531ab6a6
Submitted by rvolz.eth on 2026-11-05 UTC, 1 upvote, 0 comments

## Summary

Laso Finance reports that agents still fail to operate as founders, based on interviews with Laso users, partners and acquaintances who tried building agent-first businesses, plus Reddit feedback. The most cited blocker, at 53%, is not knowing whether an agent's action happened, such as whether an ad posted or an email was sent. Login, two-factor and texted-code requests came next at 31%, followed by money problems at 28%, including card fraud blocks and no safe way to hold funds. Laso says it has removed the first blocker with an agent-native financial suite and is now giving its agents email addresses, phone numbers for 3DS requests, and ad-buying abilities.

## Article

There is a lot of excitement about agentic commerce. Every fintech has joined the fray.

The technology, talent and demand (from builders) are there. But to quote Fermi:

Where is everybody?

Muse and Instinct are finding traction for consumer usecases. However, we shouldn't be excited about agentic commerce because it can buy items for your home off Amazon.

We should be excited because of the businesses that either wouldn't or couldn't be built from human labor alone. Agents should be out-executing human founders.

However, they're not here yet. Where are they?

Research

We conducted multiple user interviews of Laso users, partners, and acquaintances who have attempted building agent-first businesses. In addition, we solicited feedback online, mostly from Reddit.

The interviewees were building a wide range of businesses such as:

Financial services for export/import businesses in North Africa.

Digital products (e.g. wedding prep checklist) on Gumroad and Etsy.

A bar/restaurant happy hour finder for NYC.

Medium-sized commercial/residential maintenance company.

Interviews were conducted in-person, or via video call.

The questions were all variations of:

When using an AI agent to create, or run part of a business or make money, what still forces you to step in?

For such a fast-changing industry, the answers were remarkably consistent.

The Blockers

We expected the problems to be what many agentic commerce companies are working on. Issues like:

Trusting agent decisions, especially purchase decisions.

Limits on transactions.

Ability to pay.

Validating identity of the agent.

Instead, we found that the blockers are tedious, sludge work.

Not knowing whether an action happened (53%)

This is when the owner cannot be confident about the completion rate of an agent's task. It slows down the owner and more importantly, erodes trust in the agent to operate autonomously.

Issues like:

Did the ad I post go through?

Is the code pushed?

Was the outbound email sent?

Did we receive a signup?

Sometimes the lack of clarity on completion resulted in financial loss:

I had to check paypal and manually refund one guy because agent sent duplicate invoice. Happened maybe twice a week

Not only annoying, they serve a crucial function as a trigger, which is necessary for self-improving agents. More on this later.

Logins, two factor and texted codes (31%)

Maintaining the agents logged-in status, or completing its signup, was the next most significant blocker.

In our experience at Laso, this is the most significant blocker.

Examples of these issues are:

Can't connect agent to Google Ads.

Connected to Meta Ads, but appears to require human to submit creative.

Connecting Vercel MCP requires OAuth flow, but agent was on VM where human can't easily access agent's browser.

Agent can't create an Instagram.

Google Seach Console is blocked to agents.

Some direct quotes:

SMS verification is the main blocker, because most platforms want a code sent to a real phone number

Google Ads… every few days it asks for re-login and I lose half hour

Money: funding, cards and refunds (28%)

If the owner and agent make it through the blockers above and get a service up and running, they run into issues with their agent paying for services or receiving funds.

The bank's fraud detection still kills it every time it tries to spend more than a few bucks without me texting back YES

had no safe way to hold funds or issue refunds, so every purchase looped back to me. i basically became a manual stripe endpoint the agent pinged.

Most of the feedback was on inability to spend money. Most agents are not receiving revenue yet(!), so receiving revenue and deploying it efficiently was not an issue mentioned often.

Approvals: OK for money and irreversible steps (28%)

Today, agents are conservative regarding confirming with the owner before taking an irreversible action, such as financial transaction.

In conversations earlier this year, there was significant concern of agents purchasing too much of an item, the wrong item, or similar without authorization.

Now, agents are too conservative.

I'd want approval for spending money or anything irreversible, but not for every little click. that's just babysitting with extra steps"

When an agent does attempt a transaction, the bot-like actions often trigger a 3DS request by the merchant. If the agent's card doesn't have the agent's phone number or email attached, the human has to jump in.

3DS is important because you have to jump in to help any way. Usually only have a few minutes.

The best pattern people described is a short approval queue with the evidence for the transaction/action attached.

Reviewing and verifying the work (13%)

As a result of the reduced trust from Not knowing whether an action happened, owners felt that they had to constantly keep themselves in the loop.

They did not benefit from the speed agent founders can execute at, since the agents didn't have permission to operate autonomously.

No usable API (9%)

Some services can't be used by agents. Those failure modes fall into a few categories:

APIs that require human sign-up. (Google Ads, Instagram)

No API at all, requiring scraping. (web checkouts, Shopify)

API where a token for agent gives limited permissions (Vercel project tokens versus Vercel member).

From a user:

anything without an api key got cut… ui flows fail silently

Runs that die with no alerting (6%)

This is the second category that did not show up often in interviews, but was a very pertinent failure mode in the Laso team's experience.

Agents themselves can fail independently from 3rd party services. Examples include:

Running out of inference credits.

Virtual Machine (VM) for agent runs out of memory, storage or bandwidth.

Inference provider fails due to timeout, connection drop, or any other infrastructure error.

This is the most impactful error mode. The agent loses its ability to self-heal when it has "died".

That often results in the owner not being notified of the failure, so the agent's self-improving loop ends.

Actions from Research

Like a lot of founder work, the blockers are unglamorous. @bhorowitz captures it well in The Hard Thing About Hard Things:

There is no silver bullet that’s going to fix that. No, we are going to have to use a lot of lead bullets.

Laso Finance's north star is to enable agents to be founders.

To resolve that, Laso is now focused on fixing these blockers for Laso agents. Every Laso agent:

Gets an email address and phone number.

The email/phone is the agent's card's 3DS information, so the agent resolves 3DS requests.

Can code and publish a website (including a custom domain).

Charge for services over x402.

Purchase ads on Google, Pinterest, and more.

More importantly, as Laso agents learn new skills (e.g. how to submit a sitemap to Google Search Console without a human), those skills are submitted, reviewed, and distributed to other Laso agents on the network.

We are executing on removing those blockers one-by-one.

Lead bullets.

The Unlock

The first blocker Laso removed is a full agent-native financial suite. Today, Laso agents have the financial primitives an agent founder needs (banking, cards, on/off-ramp, onchain wallets, and all routes between the products).

Now, the clear blockers to self-improving agent founders are maintaining these external 3rd party connections, and keeping the agent's autonomous loop alive.

The key to agents being impactful is their ability to operate and learn sustainably without human input. They often fail due to the reasons described above.

To quote @different_mj:

We started with humans spending tokens. Now, tokens are spending tokens. Humans still spend the majority of online dollars. Soon, tokens may spend the majority of online dollars.

For that to happen, agents can't just spend dollars. They need to earn them.

To earn dollars, the issues described above need to be solved. Laso is removing those barriers one-by-one.

We are looking for agent-native pioneers who want to be part of this sea change. Or who have attempted to build an agent-native business.

If you're part of either group, DM me for an invite to Laso Builders.

Appendix

Reddit Feedback

AgenticAI

LLMDevs

AI_Agents

Automation

AgentsOfAI

AiAgents