Transcription
Okay, I'll say it. Building agents is the new SAS. I mean, we saw billions of dollars of value creation during the SAS era. People, founders, 21 years old, 24 years old, you name it, come up with SAS ideas that changed their lives. Of course, not everyone was successful, but this is a new wave that's happening. And I want you to understand what this wave is, how you can take advantage of it, how you can build startups in this space. And in this episode, I'm basically going to clearly explain this entire agents is the new SAS opportunity.
By the end of this, you're going to understand the playbook. You're going to understand how to find the niche. You're going to understand how to pick the workflow, how to, you know, build your first agent, how to prove it works, how you can package it like SAS, and how you can basically sell these agents as labor. And it's really important because that that labor point because labor is a multi-trillion dollar market, right? So I think why people say agents is bigger than SAS is because the total addressable market for agents is just way bigger. You know, it's human capital. And I wanted to just do an episode where I explain everything step by step just clearly for people who are interested in building agents, who are who are interested in selling agents, and who are, you know, just want their creative juices flowing around this. This is for anyone who wants to build a business or who just want wants to be more productive. This episode is for you.
So the first thing you need to understand is the product is the job. So, you know, the mental model I have for this is SAS sells software, agent SAS sells work. A normal SAS product says, "Here is a tool a team could use, basically." Um, but an agent SAS product will say, "Here is a job your team no longer has to do by hand," and then you're selling that service. So it sounds like a a small change, but it's a it's a really big mindset change for the customer and for you, the person who's going to be building it.
So take, uh, restaurants, for example. You know, a restaurant wakes up thinking, "Our phone is ringing during, you know, dinner time, um, and the host is busy, right, dealing with, you know, seating people. People are asking the same questions. We're missing reservations, and private dining calls are getting lost." So, they're basically missing out on lost revenue. This is why a company like, you know, Slang AI is interesting. Basically, it's an AI superhost for restaurants. Not affiliate, I just think it's an interesting, you know, example. Uh, it answers inbound calls. It handles guest questions. It manages reservations. It routes VIPs. It alerts staff about high priority topics like private dining or guest complaints. And it integrates with systems like Open Table and Yelp and stuff like that.
I'll give you another example. Home services, a plumbing company, HVAC company, a roofing company, pest control company, miss calls, you know, jobs are booked, customers followed up with, dispatches are less overwhelmed, more revenue from the same demand if they integrate something that picks up their phone 24/7. You know, an example of a startup that's doing this is Same Day. So they focus on home services and then they basically sell these AI, you know, dispatchers, sales agents, receptionists that answer calls, respond to texts, book jobs, reschedule. Again, the product is the job.
These examples matter because they're just understandable. And so when you're thinking about coming up with an idea for a startup here, think about, "I handle this one annoying job better than a junior employee, faster than an agency, and it's cheaper than adding headcount." That's the mental model I think that you should be thinking about.
The second step I have here is pick a workflow with a paycheck attached. So, you know, how do you find the right agent idea? You want to start with the paycheck, right? If people are already paying for the work, they're paying an employee, they're paying an agency, a receptionist, a coordinator, dispatcher, they there's an opportunity to sell that service cheaper and then unload some of that work that that person is doing to do more highly creative creative work. So a good agent workflow has five traits.
Um, the first is it happens all the time. So daily is good, but hourly is better. Uh, you know, every inbound lead, uh, every call, every tech ticket, every quote request, every appointment, every order, every maintenance request. Second, it has to have a clear finish line. Uh, the job got booked, uh, the ticket got categorized, uh, the refund got approved, uh, the vendor got scheduled. Uh, the customer got a useful answer. Uh, third, uh, if it touches already software. So if it touches Gmail or Slack or Shopify, HubSpot, Zenes, Stripe, um, agents need these tools so they can use it, and they also need the context that they can read. The fourth is the edge cases are actually really annoying, um, but they're learnable. So if a workflow is too basic, um, basically the problem is basic automation, you know, Zaps and stuff like that can do it. Um, if it's pure human judgment, uh, the first version will break. So the sweet spot is repetitive work with enough judgment that AI can help. Uh, the fifth is the buyer can feel the loss. Missed calls, slow replies, drop leads, uh, empty calendar slots, uh, expensive humans doing low-value, uh, coordination.
So, if this is, you know, you know, if you want to start a business here, here, what's a first rep to get started? Well, pick one niche and write down 20 jobs people complain about. Um, if it's roofers, you know, maybe it's missed calls, uh, financing questions, insurance paperwork, appointment reminders. If it's med spas, it's maybe lead qualifications or no-show recovery, membership upsells. Um, if it's Shopify brands, maybe it's, uh, returns, uh, exchanges, uh, wholesale lead follow-ups. And then you want to score each job on five things. Um, how often does it happen? How expensive is the pain? How easy it, how easy is it to know when the job is done? Uh, what tools does it need access to? And who already owns the budget? And that is where agent SAS starts. So you want to start with the job that has a paycheck attached. Super, super key point here.
Um, once you've done that, you actually, and this is this is the thing that a lot of people miss, you you actually want to shadow the human before you build. So once you find the job, right, you you've identified what the job is, before you like prompt, before you start coding, just shadow a human being who does the job. Uh, I know a lot of founders want to skip this. Um, but it's usually what's going to help you get, you know, an unfair advantage, an insight into building an agent that is high quality. So, just watch someone do the job, 10 to 20 jobs. Ask them to screen record it. Um, ask them to narrate what they're doing. Uh, ask them what makes a case easy. Uh, ask them what makes a case weird. Uh, what did they check before they make a decision? And ask them, you know, where the mistakes happen. Basically, you're looking for the real workflow. Um, and even if you've done these tasks before, having having it, you know, fresh in your memory, is just going to be super, super helpful when you're building out this.
So, you know, for example, a restaurant host, if they answer, "What time are you open?" Uh, the real workflow is actually deeper than that. Um, they know when the kitchen closes, which tables are good for strollers, uh, when the patio is closed, how to handle a VIP, and, you know, when to route to a private inquiry. So the detail is the product. Um, and when you're specking out your your agent, you know, I think it should have seven key parts. What wakes the agent up? What context does it need? What tools can it use? What is it allowed to do itself? Where does it need approval? When should it escalate and and bring a human in the loop there? And what does success look like? And if you understand all those things, um, you're, you're not just going to build an agent slop, right? We want to build agents that are high quality, that are exceptional, that do the work as well, if not better than humans, and way more consistently, and that's what people are going to pay for. So, uh, really helpful to just sort of internalize that.
Now, you're going to want, uh, step four is you're going to want to build the smallest useful agent. You know, I call it the minimal useful agent, the MUA. Most people hear agent and imagine a fully autonomous employee. Um, and that's how you you get these demos that you see on Twitter. Uh, and they don't really work, and it ends up being a bad business. Um, you want to start smaller, actually. So, there are four good first versions. The first is a draft and approve agent. So, it reads context. It drafts the reply, the quote, the summary, or the next step. A human approves it. Uh, so this is great when there's workflow risk, um, maybe there's creativity involved, uh, approval processes, that that sort of thing. The second is a triage agent. So, it classifies inbound work and routes it to the right place, maintenance requests, billing issue, or, you know, refund, that sort of thing. The third is a coordinator agent. So, this goes between systems and people. So, it'll check availability, it'll send reminders, it asks for missing info, um, and it keeps the work moving, basically. Um, and the fourth, the fourth is the bounded action agent. So, it can do a specific thing under clear rules. You know, book an appointment, send a follow-up, you know, process a refund under $50. These are the sorts of things you'll you'll now see. You'll notice it in the wild. Like, for example, with Uber Eats, you order something and your salad doesn't arrive, automatically you'll get the refund, um, based on an agent that will do it. So that's basically the the ladder here. Uh, when you're building the the minimum useful agent, draft, triage, uh, coordinate, and act.
I was looking at Anthropic's agent guidance, uh, recently, and they made a really, uh, important point, a really important simple point. They said, "Many agent problems should start actually as workflows." So a workflow follows a predictable path. An agent decides more dynamically. Founders should earn autonomy by starting with a predictable path and adding judgment only when it creates value. So if I were building this, I would start with one workflow and one promise. For example, "We answered missed calls for roofers and book qualified jobs," or "We triage maintenance requests for property managers and schedule the right vendor." "We handle reservation calls for restaurants and alert staff when a human should jump in." That's enough. That's really enough. And one workflow that works enough is is good enough for day one. And it's going to help just like build confidence, not just in you that it's working, but also in this customer. Cuz keep in mind, people are buying agents for the first time ever. So they they want to, they don't want all of it at the same time, especially if you're not Microsoft or you're not Salesforce.
Step five in building an agent is the product wrapper is what makes it a SAS. So you don't want to just build a cool automation. And I think what separates a cool automation from a real agent-first SAS product is the agent does the work, but the wrapper creates the trust. So customers actually need to see what happened. They need to see logs. Uh, they need approvals, um, controls, handoff rules, that sort of thing. And, uh, they need a way to test the agent before it goes live. They need to know basically why the agent did what it did. So that wrapper is the SAS, and the agent actually lives in the phone system, the inbox, the Slack channel, CRM. So the dashboard can be really simple. Um, but the customer, customer actually still needs the control room. Um, at least that's what I'm noticing is working right now.
So for a restaurant phone agent, the control room might be, you know, call summaries, or, uh, reservation outcomes, missed human handoffs, that sort of thing. Um, for a property maintenance agent, maybe it's, uh, you know, tickets that have been created, vendor routes, or tenant updates, owner approvals, that sort of thing. Um, and that's really why eval matters a lot when you're building an agent-first, uh, business. Um, because before you promise autonomy, you want to create a basically a small test, uh, a test set is the best way to think about it. So, you know, take 50 real examples of the job and then mark the right answers, like 50 calls, you know, 50 leads, uh, maybe, you know, 50 maintenance requests, depending on what it is. And then you want to run the agent system against them. So, did it classify the problem correctly? Did it ask for the right missing information? Did it use the right policy? This is what's called an eval, right? Your eval set is, you know, it's basically like the gym. You know, every time you change the prompt, the model, the tools, the workflow, the agent goes back through the gym and and is able to, you know, basically know what's good and know what's bad.
It's also like low-key a really good sales asset because imagine telling a property manager, you know, "We tested this on, you know, 50 of your old maintenance requests. It routed 42 correctly, flagged six of them for human review, and made two mistakes. Here are the two mistakes, and here's how we fix them." So, it just helps you actually build trust, uh, especially with people who are like who own boring businesses, but who can actually really use some of these agents. When you, when you're just transparent and open about these sorts of things, you're going to build a lot of trust with them.
So, uh, the sixth step, uh, on building an agent is you're going to want to sell the pilot like labor, and then you're going to want to productize it. So, the fastest path is usually a pilot where you manually do the work with AI and then you productize the repeated parts. I'll explain what I mean by by that, but, you know, basically, I would start with three customers in one niche. So the same niche, same workflow, same pain, and then you sell the outcome. You actually want to be like very, very, uh, constrained on this, or else you're going to get bad results. So if you're selling the outcome, it's, "We will answer and qualify your missed calls." "We will triage your maintenance request." And then you charge a setup fee and a simple, like, monthly fee. Um, you can keep the pricing just like easy to understand, and then you can add usage or outcome pricing once you understand the value. I'm a huge believer, if you're, you know, listened to the Startup Ideas podcast here, and, you know, you're a subscriber here, you know, I I really believe that outcome pricing is the future of how a lot of these agent-first businesses and software is going to be priced because, you know, it just makes sense, like, you know, the customer only wants doesn't want to pay for another seat for something.
So, uh, but don't just jump there, uh, initially. Um, so you'll get there, and you have to have patience with it a little bit. You know, the the pricing model, like how do you think about pricing, uh, something like this? You know, maybe it's like a $1,500 setup, $1,500 setup, and then like $1,000 a month for one workflow. Or maybe it's like $2,000 setup plus $30 per qualified appointment. So that's more the outcome-based eventually. Or $3,000 a month up to 500 handled tickets. So the exact price actually matters less than the learning. Uh, you basically want to find out what the customer values, and then, uh, where the agent breaks, uh, what needs approval, and what would they miss if you took it away. Pretty much the most important question there is, and then you build the product around the repeated pattern.
So, if every roofer needs the same emergency call script, service area check, financing question, an estimate follow-up, boom, you have a product. Um, if every med spa needs lead scoring, consultation booking, no-show recovery, um, post-treatment follow-up, boom, you have a product. So, this is agents SAS. You know, you earn the software by doing the work first.
Step seven is distribution. You know, you're going to want to get, you need obviously, you need people to hear about the thing you're building, hear about the agent, see it, and be like, "I need this." And how can you do it? What are some tips to to be thinking about this? Workflow tear-downs is what I'm seeing in real time that's working really, really well. So, show the old way of doing a process, right? A call comes in, nobody answers. The customer calls the next company, or or the CSR answers, asks five questions, checks the calendar, checks the service area, books the job, you know, you know, writes the notes, sends a reminder, and then forgets to follow up. And then the heart of the owner, of the manager, just absolutely breaks. And then you show the agent way. A call comes in, an agent answers, it asks the right questions, it checks the service area, it checks the urgency, it books the appointment, it updates the CRM, it sends the confirmation, it flags edge cases for a human. That's the type of content that works because the manager, the executive, the owner feels that pain. And, you know, you want to be in the business of selling painkillers, not vitamins. And that's why I love these building agent businesses because you are, you know, if you do this right and you go through some of these steps and and these frameworks and this playbook for coming up with an idea, building an idea, you're building something that's really, really helping people. Like it is very, very valuable to them. And the only way to really, really get them to book a call or do something is to show them, sort of make fun of the old way and then show them the new way. Pick one workflow, make the internet associate you with it. And, you know, the way I think about it is you want to make the checklist, you want to make the benchmark, make the tear-down, make like 50 examples of this workflow post, and you're going to be in the content game. You know, a lot of people listening to this, you know, don't want to create content, and I totally understand, but, you know, you have to create content around this because it's just the, you know, there's just an opportunity in creating content. Um, and then you can use some of those assets that are starting to work and then put paid, uh, paid ads, uh, around it. So, you know, distribution, think about these these tear-downs, think about poking fun of the old way, think about creating memes around it, uh, and then creating content, picking the winners, putting paid ads against it. I would suggest focusing on one platform to start. Um, and, uh, you know, I'm going to be doing more episodes on just how do you how do you get customers to your your vibe coded apps and stuff like that in the future. So stay tuned for that. Um, but just, you know, how I'd be thinking about it.
The last step, you know, the zero to 100 plan, like if I was starting this from zero, you know, I feel like people would want to know what would be my plan. Um, so I'll give you my, you know, call it four-week plan. Um, my first day, what I would do, like how to build a business. How do you build an agent business in 30 days? Okay. Day one, we pick a niche where missed work costs money. We talked about this. Home services, property management, you know, insurance agencies. Day two, I would interview 10 operators, ask them to screen share the workflow, keep the calls as research, and just watch them. You can pay them for this, by the way. Day three, pick one workflow with frequency, pain, software access, and a clear success metric. Day four, write the agent spec, trigger, context, tools, rules, handoffs, eval. Uh, day five, run it manually with AI. You know, use Claude or ChatGPT and copy and paste the context, draft the output, uh, ask the human to approve, and you are testing whether the AI helps before you build the software. Day six, build that smallest useful version. Uh, draft and approve or triage is usually enough here. Day seven, create the eval set from 50 real examples. Week two, sell two pilots in the same niche. Week three, add the product wrapper, right? The logs, the approvals, the settings, the analytics, the handoffs, and use AI to actually build that, you know, software. I'd probably using something like Claude Design. And, you know, assuming that Fable is live, using Fable to actually code it up. Week four, I'm publishing workflow tear-downs. I'm turning the pilots into proof and I'm doubling down on my content strategy. So, you know, while, you know, I I should have mentioned this, but while I'm doing all this, while I'm thinking about building this business, I'm building an audience all the way throughout. And by the end of week four, I have formats that are working. I know where to double down. I know where I can spend money, paid money to acquire customers. And that's when I, you know, my second and third month, I'm trying to understand what my LTV is. You know, what are the channels that are working, where can I double down on it, and I keep going.
I think that's the episode. You know, basically where I'm at is I think agents are the new SAS because software is moving from "help me do the work" to "do the work with me." And I think that a lot of people are just they're not partaking in the shift. They see the shift, they understand the shift, you're listening to me, and you're like, "Yeah, of course." But they're not building, they're not building agent-first businesses. Um, I think the opportunity is basically to find the smallest painful workflow that repeats all day, in a niche that you understand, and make it disappear. You know, answering a phone, booking a job, triaging the ticket, uh, updating the system, uh, escalating that weird case. Um, and I think there's a ton of money to be made here and a ton of value to be created. So, if you're watching this, you're listening to this, and you're thinking, "I want to build with AI," well, start start with the job. You know, find the job, shadow it, spec it, run it manually, build the smallest useful agent, sell the pilot, then productize the repeatable parts. This is how you build an agent that people will pay for. And this is why I think building agents is the new SAS. Hope this is helpful. Hope this got the creative juices flowing. People charge thousands of dollars for this type of thing, but it is free, like always, on the Startup Ideas podcast. Let's I'll see you in the comment section. I'll see you, uh, let's talk and let me know what you want me to cover next. If this is helpful, and I read every single comment on YouTube, I appreciate every single like and subscribe if you want more of this stuff in your feed. Thank you, and I'll see you next.