Custom Software Development in the AI Era: Hype vs. Reality

AI-powered custom software development is evolving fast in 2025, but does it spell the end of low-code development? After all, AI agents can now drop thousands of lines of code from a voice memo you recorded while waiting for a smoothie… right?

Eh, not so fast. 

Panicked chatter on the socials aside, AI really is changing how companies develop custom apps — just not in the way that makes the hypebeasts freak out.

Recently, I sat down with Paulina Rios-Maya, who heads up industry relations at EM360Tech. We took a deep dive into the ways that AI is shaping custom software and internal apps in 2025 (check out the podcast here or on Spotify.) 

This post is the first in a series discussing the changes we’re seeing as more and more companies look to leverage AI in their operations. The TL;DR is that AI is causing a huge shift in custom software development, but…

• It’s not doing everything;

• Humans are still central to business processes; and

• Low-code tools may actually become more valuable through the transition to AI-powered apps.

Here’s what we’re seeing in custom software and low-code development as we plunge deeper into the AI era.

Let’s start with the bad news. Despite all the hubbub, no one’s ready to hand off critical parts of their business to an AI agent just yet. A few routine tasks, sure. But authorizing refunds? Approving loans? Forget it. Everyone still wants human oversight when there’s real money on the line.

But AI is changing the custom tools that companies use to do those things. It’s just not ready to act autonomously. Rather, it works better as an assistant right now — handling the nitty gritty and checking in with its supervisor before making critical decisions.

Today, a lot of corporate work involves checking multiple sources of information and synthesizing them together to make a decision. This is a process that AI assistants are set to transform.

Say you’re a customer support rep for BestBuy and someone wants to return a TV. Before you can decide “yay” or “nay,” you need to answer a bunch of questions:

• “When was the TV purchased?”

• “Does that date fall within our refund window?”

• “Do we categorize TVs as large appliances?”

• “Do large appliances have different policies from standard merchandise?”

• “Was it on sale?”

The data to answer those questions may be stored in three or four places: corporate policies, CRM software, delivery receipts, etc. Companies currently use low-code platforms for exactly this reason. They make it easy to pull data from multiple sources into a “single pane of glass,” so employees have the information and context to make a good decision (like authorizing or denying a return). 

But this process looks very different with an AI assistant.

We’ve already seen that chat is one of the most intuitive ways for humans to interact with software. More and more, the UI for internal applications is going to simplify and orient around chat. 

In an AI-powered app, you type a question or a command into a chat widget. Then, behind the scenes, the chatbot triggers a set of microapps that source data, aggregate it, and return it to the UI. 

This is great from a customer perspective: It lowers the learning curve and makes the overall experience much more straightforward, reducing the need for extensive training or consultants that help workers use the digital tools at their disposal.

In the above example, the customer support rep could ask the chatbot…

• “Check the original purchase order and find out if we’re still within the refund window.”

• “Was the TV on sale when it was purchased?”

• “Show me the original receipt, so I can confirm what you’re saying is true.”

If the chatbot is hosted in a Google Chrome extension, the rep can check the customer’s order history in Salesforce and see a rundown of their purchases. Then, they can ask questions that aren’t immediately apparent in the portal, like “What’s the average cost of returns this customer requested over the past three years?”

Overall, the workflow is much more streamlined and collaborative — more like a manager and a plucky assistant trying to solve a business problem, rather than an analyst trying to vet a request by looking at charts and figures.

AI also changes how development teams approach building the custom software their companies need. Rather than the elaborate dashboards and forms that now take center stage, a stripped-down UI focused on chat means that front-end design needs less attention — and that integrations with data sources are more critical than ever. 

Beyond writing text, developing code, and answering questions, LLMs are very powerful tools for integrating datasets. This makes them a natural fit for a low-code tech stack used to build internal custom apps because data integration is a huge reason why companies already use low-code tools in the first place.

Without AI, systems can only speak to each other through APIs — and that’s if they have compatible data. If you want to synchronize your Zendesk tickets with your Intercom messages with Jira issues in a single dashboard, you need a tool like Appsmith that has APIs with all three systems.

But AI can bridge the gap between these systems and more — taking data from one, extracting it, transforming it, and formatting it in a way another system can use. 

Take expense management. If you’re buying stuff for work, you probably get a lot of bills in your email. But expense management tools can’t read your emails and find receipts. You have to search your email, download PDFs, and upload them into Ramp or Expensify manually.

AI can do it automatically, though. It can comb through your email, find the right information, and drop it into the expense database. If configured right, you don’t have to do anything but check your bank account to make sure your reimbursement went through. 

Plus, companies don’t need to deal with a complex data warehouse or ETL process. They just hook the tools together — their email server, their expense software, an LLM, and a low-code platform to run the logic — then let ‘er rip.

“But my tool has automations and integrations out of the box!” you say.

Sure, but do they integrate easily with your internal database? Probably not. This has always been a key advantage of low-code platforms. Tools like Appsmith make it easy to synchronize data from your own private Postgres or MongoDB database with other data sources, then use the aggregated data to easily find patterns, update records, track KPIs, and perform other routine business tasks.

That advantage holds true in the AI era. 

‘Why can’t I just do everything with AI?’

It’s a huge pain to get your internal data into an LLM directly — especially in a secure, compliant, and scalable way. But low-code platforms already have those integrations ready, including for AI tools. 

Using a low-code platform with an LLM in the mix, you can easily build an app that can review, process, analyze, transform, visualize, summarize, explain, or run numerous other functions on data it pulls from your data sources — including your internal databases. You don’t need to worry about the headache of exposing your internal database directly — and you don’t need to cross your fingers and hope that a commercial vendor will build a tool for your weird long-tail use case that’s critical for your business. 

Even though AI is driving a stream of changes, a big question hangs over the whole industry:

How are enterprise-scale companies actually going to get value from it?

The truth is that the majority of AI projects in organizations have failed so far. But they’re not failing because LLMs aren’t ready or because AI is just an investor-driven bubble.

They’re failing because organizations underestimate the massive changes involved in remaking their businesses into “AI-driven enterprises.”

You can’t just drop a revolutionary technology like AI into your normal way of working and expect great outcomes. If you’re going to implement it, you need to reconsider…

• How your data is structured

• Which roles you have in your organization

• How employees think about the jobs to be done

• How training prepares workers for their new tasks

• Which SOPs to follow

The AI packages that directors and C-levels are buying off the shelf are delivering no value because companies are trying to apply them to “business as usual.” That’s not going to work. Companies need to work through the changes necessary to use AI effectively — and it’s often a lot.

This is actually an area where low-code tools can help. Enterprises are understandably iffy about remaking their entire business from the ground up. Even if the long-term gains are enormous, short-term pains and the stakeholders who suffer them can cause inertia.

But tools like Appsmith offer an easy way to introduce AI functionality without taking a wrecking ball to your business SOPs.

If you’re already using Appsmith to manage customer support tickets, identify upselling opportunities, or forecast churn risk, you can plug an LLM into your existing process and start reaping the benefit. (If you’re not, it’s easy to get started.)

Your team can experiment with a chat interface and learn to rely on it more and more as they become comfortable. Your organization can start small, adding AI into the tools and processes you already have, before jumping into the whole new world of AI-dependent enterprises.

We’re already seeing this play out at small and medium-sized businesses. Customer support teams at these organizations are starting to adopt AI tools to handle basic tasks, such as responding to L1 messages like “Why is my internet so slow?” or “What’s the status of my ticket?”

It’s no surprise that SMBs are driving the change: They’re more fault-tolerant than huge enterprises. But these early successes are AI’s foot in the door — and they showcase a practical way to deliver value across an entire business function, instead of just drafting a one-off email.

If enterprises learn from these experiments, we’ll finally see the business transformation that the AI vanguard has been forecasting. But it all starts with AI vendors offering practical solutions to business problems — just like low-code platforms have for years.

If you want to see how we’re building the future of custom software with AI, we’re looking for developers to beta test our upcoming AI Assistants feature. If you have some experience with commercial AI solutions and you work in support, customer success, or sales, join our waitlist for early access.

AI isn’t here to replace developers — it’s here to make their lives easier. It automates the boring stuff, pulls in data from 12 different tools, and makes custom app development less of a headache. Instead of spending hours wiring up integrations or digging through documentation, teams can focus on building useful stuff faster.

Low-code already makes building apps easier — AI just cranks it up to 11. Instead of clicking through a hundred dashboards, AI assistants let you ask for what you need in plain English. “Pull up that sales report,” “Show me last quarter’s refund data,” “What’s the quickest way to do X?” — done. It’s like having a turbocharged intern who actually listens.

Not exactly. AI can write code, sure. It can even automate workflows and fetch data. But when it comes to making judgment calls — especially the kind that involve actual money — nobody’s handing the keys over to an AI agent just yet.

Humans are still the ones approving refunds, green-lighting loans, and making sure critical business logic doesn’t go haywire. AI is great at fetching info, summarizing data, and handling the busywork, but it still needs a human in the loop to make the final call. 

You know those tools that almost talk to each other but not quite? AI fixes that. No more duct-taping APIs together — AI-powered integrations can grab data from one system, clean it up, and feed it into another without the usual mess. That means less manual work, fewer spreadsheets, and faster decision-making. Less clicking, more doing.

Because no one wants to rewrite their entire tech stack just to test an AI feature. Low-code platforms let companies plug AI into their existing workflows without breaking everything. Need an AI-powered chatbot? A smarter dashboard? Low-code lets you build, tweak, and experiment — without waiting six months for IT to approve a budget.

AI isn’t a magic wand — it’s a tool. And like any tool, it only works if you know how to use it. Most AI projects flop because companies expect instant results without rethinking their workflows, training their teams, or structuring their data properly. AI success isn’t just about having the right model — it’s about having the right plan.

Start small, keep it practical. Don’t try to “AI-ify” your whole company overnight. Instead, add AI to the tools your team already uses. Let it fetch data, automate simple tasks, or make recommendations — stuff that saves time without breaking processes.