I have a fun job, I get to work on bleeding-edge stuff all the time and naturally, AI has dominated my projects for the past few years. I have tested nearly all of the major models and other Python toolchains like PyTorch, langchain, etc...
One major problem with most AI projects, they usually end up being a value add-on to existing products and not a standalone product on its own, except for "wrapper" services.
Yet, every other day, there's talk about AI taking over the world. Here are some real-world projects you can build to make revenue from agents.
AI accountants
One of the most annoying things for small business owners is having to deal with accounting, especially for SaaS founders who can't shell out hundreds or even thousands of dollars in accounting fees in the first few years.
An agent to simplify this process would be nice. Here's what the agent can do:
Connect to banking APIs and scan transactions, categorize them into the relevant taxable types.
Accept input from an email/text message or even something like WhatsApp. Think of a plumber out in the field, it's cumbersome to make invoices on a mobile or tablet device. The agent can parse text messages and generate an invoice, and then email that invoice on approval. Furthermore, when the customer pays, it can associate the invoice with a transaction from their bank account.
Based on all the data in the system, generate balance sheets, tax submissions, etc... Off course, there needs to be solid checks and balances to ensure the AI doesn't do something weird. Probably working with an accounting firm to audit these would be a good idea.
AI Receptionist
Again with the plumber example, these artisans are constantly on the road, and for a one-man-band or small team, having a dedicated receptionist is not always feasible, thus, the plumber himself/herself generally will also manage calls which can be painful, not to mention cause them to loose some leads when they're busy on a job.
You could wire an AI system to the telephone line and let the agent book appointments, take messages and also provide information too.
This should be fairly easy to do with Elevenlabs and Twilio.
AI web designer
SaaS founders, small businesses, and other professionals usually need a quick landing page to showcase their app or services. These websites are mostly static and don't need to be overly complex; usually, a one-pager with testimonials, contact forms, maybe a few photos, and a blog section should suffice.
Using Tailwind, you can craft some "seed" website templates and then build an agent that uses a web search to source images from sites like pexels.com and get inspiration from similar businesses.
The agent can then interactively go back and forth with the user to generate the design and make any changes needed.
Conclusion
AI agent frameworks frequently contain unnecessary complexity, with examples such as n8n, LangGraph, and similar tools making implementation more complicated than it needs to be.
Now, yes sure, use the right tool for the right job, so I am not saying don't use these, they do serve a purpose.
The reason I point this out is because I see more often than not, developers are building agents because it's the hype thing right now, which is fine, it's a learning experience.
Although similar to a recent Satya Nadella interview, all these AI tools are cool and fun, but at some point they need to impact GDP growth, and as a developer this could be a great oppturnity to monetize your skills and maybe even build your own startup.
Essentially, agents don't have to be overly complex; you can pretty much just steal my simple example here and just build a small MVP in a couple of days.
Hopefully this article gives you enough of a start to actually build something that's commercially viable. I will expand with more ideas here when I get a chance.
🙏 Please comment below if this is useful to you or if you would like me to go through some practical examples on how to build these agents in a future article.
