Published by Tenth Gear Consulting

Creating your own app? Remember, you don’t know what you don’t know.

App creation and experimentation by non-technical users had been slowly growing. AI-assisted coding and deployment allowed deep domain experts the freedom to explore and create. You just talk to AI and it makes and publishes the app, who needs devs?

Is it really that easy?

The 80% problem

Reading through support forums reveals a pattern - user creates and publishes the app, then runs into an odd problem that AI agent cannot solve (or in some cases, makes worse).

Because they lack the technical background, the helpful suggestions posted by others are not really helpful and the whole experience goes sour quick.

But that’s not the worst that can happen. A non-technical user who shared their experience building and publishing the app found himself hacked, with API key stolen and credits for AI usage calls used up. The situation was caused by a significant security holes created by code generator and security keys exposed in plain text.

And while you can ask the agent to correct the issues, you need to know what they may be to even notice the problem and ask the right questions.

You don’t know what you don’t know.

Of course, even technology companies have issues with their software full of bugs and security holes. But they are well equipped to manage the fallout, handle risk and have the remediation know-how.

Experiment, find and test out use cases - but before releasing your code into the wild, find a technology expert to advise you. Better yet, have a sidekick with some level of technical background to help you with best practices.

After all, you wouldn’t be changing electrical panel by yourself after watching a single video.

Don’t lose your progress

A good example of software construction rule is to save the incremental changes to your app in the repository - so when AI agent goes rogue and wipes out the code that was working - or a new model gets released and decides to change everything, you can still go back.

Costs will increase

Since you have to pay for AI usage, it’s important to be realistic about your costs. Getting to 80% may cost relatively little, but as your app grows, the cost will increase exponentially with size and complexity.

Be realistic about longevity

“No worries about 80% issue, I’ll just hire someone to fix the last 20%”.

Realistically when a human engineer look at your AI generated code, edits would be required. A typical reasons may be overcomplicating the code logic which makes maintenance expensive, security gaps and badly performing algorithms. In some cases it may be easier to rewire entire sections from scratch.

So treat your app development as a real-life working prototype or you will be disappointed when taking it to finish takes longer than planned.

How to wrangle data

It’s a given that any organization regardless of size or industry would have terrible data. It accumulates in the dusty corners of your daily operations and while everybody knows that something needs to be done to make it useful, it just feels like an impossible task.

And AI pressure does not make it easier.

Here are three practical steps you can take this month:

Standardize your terminology

If your company is over few years old, I guarantee there is at least one piece of data with two different labels. You develop internal slang or perhaps things were named historically and that name no longer makes any sense. Having consistent labeling worth the effort:

• You will help new hires to understand what they are working with faster

• You will make it easier for analytics

• You can find duplicated and outdated data

Build a clean pipeline

It may not be always possible to fix “bad data” - legacy software, complicated processes that nobody wants to touch, things outside of your control like vendor feeds.

Instead of trying to “fix” the data at the source, create a brand new “clean” data feed from your existing sources, preparing it for the modern needs.

Data Classification

If you don’t classify your data, start small. Even basic three category classification significantly simplifies data governance activities.

At very least, define these categories:

• Source of truth (most authoritative version)

• Security level (public, internal, confidential, restricted)

• Retention requirements

Running an e-commerce business? Pay attention to new image capabilities.

I remember the days of automating the product photo editing: removing backgrounds, cleaning up lighting.

Then came AI models. Adding multiple product photos allowed to swipe backgrounds and scenes - it took multiple steps but was so much quicker.

And now the recently released OpenAI model can do this in one shot.

The affordability of these automations unlock new possibilities for your product media. Generating uniquely customized scenes, seasonal variants and running A/B tests on imagery will continue to get easier. A truly innovative ideas may finally be implemented on scale.

Start planning an experiment to test how different visuals may affect the customer’s response.

Watching: WordPress turbulence continues

Automattic, the commercial arm of WordPress and major contributor to open source project had announced 16% layoffs among the legal battle with WPEngine.

Why it matters: the instability of last year trouble in WordPress community shows no signs of settling down. As we suggested before, it would be prudent to investigate alternative solutions as part of business continuation process.

This month we asked:

What came first - Investing in Tech or SMB Growth?

We dig though Salesforce SMB trends report to answer this question

Is innovation just for startups?

We looked at unique advantages small businesses have.

Thank you for being a subscriber!

Have a question? Reach out anytime - all emails are answered.