← All articles

Where software is heading: nine lessons from building HotLead

By Darryn Mountfort · · 11 min read

AI · software · MCP · HotLead

I have written software for more than 25 years, 16 of them as a hands-on CTO. This year I started to wonder whether that still counted for anything.

The worry is simple. Anyone can now describe an app to an AI model and get working code back. Software companies lost about a trillion dollars of market value in a few weeks early this year, on the fear that AI agents would replace the tools people pay monthly for. If anyone can build anything, what is a developer for, and what is a software product worth?

I didn't answer that by reading about it. I built a product. HotLead is a marketing app for local service businesses: landing pages that capture leads, social posts, reviews, and a phone alert the moment a lead arrives. Building it, and watching real customers use it, taught me nine things about where software is heading. Each one comes with the moment that taught it.

1. Code is no longer the moat

For most of my career the hard part of software was building it. That scarcity protected a lot of products whose real value was only that they existed and would be expensive to copy.

That protection is gone. I started HotLead on 8 October. Thirty days later it had landing pages, an AI writer, a social calendar, reviews, appointments, billing and paying customers. That is roughly 194,000 lines of TypeScript and 61 database changes, written with AI doing most of the typing and me making the decisions.

I was proud of that for about a day. Then the obvious follow-up arrived: if I can do this in a month, so can my competitors. A long feature list no longer keeps anyone out. Everything else in this article is about what does.

2. Your software has to be operable by AI agents

This is the lesson I'd put first if I had to pick one, and I didn't plan it. A customer taught it to me.

A few weeks after launch I noticed one account doing an unusual amount of work: 11 landing pages and 53 social posts in a short time. I asked the owner how he was managing it. He said he wasn't. He had told his AI assistant, Grok, to log in and do it for him.

The assistant was clicking through screens designed for a person's thumbs. It worked, but badly. Posts went out without tracked links, so we couldn't tell which post produced which lead. Some results were fragile and some were wrong.

I had two choices. I could block it, or accept that this is how people will use software from now on. I built a proper way in:

  • An MCP server. MCP (Model Context Protocol) is a standard way for an AI assistant to connect to a tool. Any assistant that supports it can now use HotLead directly.
  • The same features as a plain API, for assistants and scripts that don't use MCP.
  • A guide written for the assistant, explaining how the system gets leads and how to work with it properly.
  • Access the owner controls. Each owner creates a key in settings, can revoke it at any time, and can see the assistant's recent calls.

The assistant can now read the business profile, build and publish a landing page, write and schedule posts, choose photos and ask customers for reviews.

The wider point is that for a growing share of products, the user is no longer a person on a screen. It is that person's assistant. If an assistant can't operate your product, your customer will move to a product it can.

3. The rules must live on the server, not in the instructions

Connecting your product to AI assistants will soon be ordinary. The harder part is what happens after you connect.

An assistant is fast, tireless and confident, including when it is wrong. My first test proved it. I connected ChatGPT to HotLead and asked it to publish a post. It did, with no picture. On Instagram that fails outright, and a picture can't be added once a post is out.

A written guide asking assistants to behave is not enough, because an assistant can ignore a guide. So the rules are enforced by HotLead itself, whoever or whatever is asking:

  • A post with its own link in the text is refused. HotLead adds the tracked link, so every lead can be traced to the post that produced it.
  • Scheduling a post with no picture is refused unless the owner has confirmed that is what they want.
  • A landing page with unconfirmed facts still marked as placeholders can't be published.
  • Each key has daily limits, so a confused assistant can't flood a business's page.

The AI writing inside HotLead follows the same standard. It never invents testimonials, statistics or results. Anything it can't confirm from what the owner said is marked for the owner to check.

Opening the door to assistants was quick. Making it impossible for one to damage a customer's reputation is the part that takes care, and it is where I think a lot of quickly built software will fall short.

4. What still protects a business

If code doesn't protect you, four things still do. None of them can be generated from a prompt.

  • Distribution. Who already trusts you, and who will introduce you. My first paying customer became a referral partner and wants to introduce HotLead to other business owners in his industry. No competitor can copy that relationship.
  • Data nobody else has. HotLead records how every post performs on each customer's own page, and compares each new post with what is normal for that page. After a few months that history is worth more than the code that collects it.
  • Being part of a real process. Software that sits inside how a business actually earns money, such as answering a lead, booking the visit and asking for the review, is hard to remove.
  • Judgement. Knowing what to build, for whom, and what to leave out.

I learned the last one the slow way. HotLead began as a tool for one estate agency, then I decided it could serve every kind of business, and I couldn't choose. The evidence chose for me: my first paying customer, my partner and my best results were all in property. I stopped arguing with it.

5. Charging per user is on its way out

Most business software charges per person who logs in. That model assumes people do the work. When an assistant does the work, there are fewer people logging in, and the price stops matching the value.

What replaces it is charging for work done or results delivered.

HotLead has never charged per user. It charges a flat monthly fee that includes an allowance of AI credits. Every AI action, such as writing a post or building a page, costs a fixed number of credits. The price of a credit is mine, and isn't tied to what the AI provider charges me, so customers get a stable price even when model prices move.

I am also testing the next step with one agency: being paid a share of the sales the system helps bring in. It hasn't paid out yet. That is useful to know, because it tells me precisely which part of the product still has to improve. A subscription would have hidden that from me.

6. General tools fold into assistants; specialist tools survive

The sell-off in software shares this year was not even. One mid-year tally had Adobe down 27%, Salesforce down 36% and Atlassian down 55%, and observed that the simpler the software, the harder it fell.

That matches what I expect. A general task list, a note-taker or a basic customer database is something an assistant can now do as a side effect of a conversation. Tools like that become features of the assistant.

What survives is software tied to a specific trade and the real world around it: its regulations, its payment habits, its customers' behaviour.

HotLead showed me this in its own numbers. On one estate agency's Facebook page, plain listing posts (price, bedrooms, suburb) reached tens of thousands of people. Our carefully written advice posts on the same page reached a few hundred. No general marketing tool would know that. We rewrote how HotLead opens every post because of it, and added a post type built around real, finished work.

Knowledge like that comes from being close to one kind of customer. It is why I now point HotLead at one industry first, while keeping the code underneath general.

7. Sell the result, not the dashboard

A plumber doesn't want a marketing dashboard. He wants his phone to ring, the job booked and a good review afterwards. Most business software sells him a tool he must learn and then operate, and most owners don't have the time.

I keep relearning this. My first customers get a weekly session with me, where we look at the numbers and decide what to post and what to promote. I offered it as a launch bonus. I now think it is the more valuable half of the product.

The results are what they talk about. On one agency's page, a typical listing post reached about 950 people before HotLead started posting there. Over the following weeks a typical one reached about 15,000, and clicks per post rose from about 28 to several hundred. Those were the agency's usual listing posts, on its usual page.

I am being careful with that claim. The lift began around the same time HotLead started, some posts had small paid boosts, and I am still confirming the pattern on other pages. But it is the number the owner cares about, and it has nothing to do with how many features the app has.

So HotLead is becoming two things: the app for owners who want to run it themselves, and a managed service where we run it for them. I expect the second to matter more over time.

8. Judgement and checking are the scarce skills

The job market tells the same story as the product. Indeed's hiring data shows that 71% of the growth in software job postings between May 2025 and May 2026 came from senior roles. Entry-level postings, by one count, are down 28% from their 2022 peak.

AI took over the part of the job that was producing code on request. It did not take over deciding what code should exist, or knowing whether what came back is correct.

That was my experience building HotLead. I wrote very little code by hand. My time went into decisions and checking. I kept a written log of every design decision, 49 of them in the first month, each with its reason. I kept a separate list of ideas that I was not allowed to build from until an idea had been deliberately promoted.

The checking mattered just as much. The AI wrote confident code that was sometimes wrong in ways only experience catches: a payment status read the wrong way round, a picture lost when two posts shared it. Twenty-five years of knowing what breaks in production is worth more now than it was before, because it is applied to far more output.

9. The interface is moving to where the work happens

I know many businesses that still run on paper job cards. It is tempting to think they haven't heard of the software available to them. Most have, and turned it down, because it was designed for someone at a desk.

A technician with dirty hands and poor signal will not fill in a form on a phone. Paper wins because it asks nothing of him at the moment he is working.

What has changed is that there is now something easier than paper: a voice note and a photo on WhatsApp, with an assistant on the other end that fills in the record, flags the parts used and tells the office.

HotLead already leans this way. An owner's main screen is a single list of people to call now. Contacting a lead is one tap into a WhatsApp message that is already written. When owners want a new feature, they describe it in a voice message and an AI interviewer asks only about what is missing.

The products that win over the next few years will ask people to learn less. For a lot of work, the screen will disappear into a conversation.

What I'd do with this

If you own a business, ask one question of every tool you pay for: can my AI assistant operate it for me, safely? If the answer is no, that tool is on borrowed time.

If you build software, the feature list is no longer what you are selling. Let assistants in, enforce your rules where they can't be bypassed, get close to one kind of customer, and charge for results.

If you are a developer wondering where you fit: the typing was never the valuable part. Knowing what to build and whether it is right always was, and there is now much more of that work to do.

I am building HotLead in the open and helping businesses put AI into the way they already work. If either is useful to you, get in touch.

Sources