Move slower. Build smarter.

AI can be useful or it can produce slop, is the thing. The fact that AI can render our pets as human beings is amazing. That we choose to do this, much less so.

Therein lies the crux. As Jeff Goldblum said in Jurassic Park, "Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should". Replace "scientists" with people and dinosaurs with "me as a muppet" and we have arrived at today.

What you need to do is slow down, stop looking at what it can do for you and your customers, and instead look at what it should be doing. For instance: language is what makes humans efficient, cooperative and capable of advancement. The fact that we can now use it to interact with computers and from there almost anything else, is remarkable. "Chat attached to FAQ" is meh. "Goofy Avatar Generator" in an accountancy Saas is cringe. "Chat attached to the client's data as a Business Intelligence Engine" is benefit.

What harm?

Agentic AI has collapsed the engineering center. What was once the most expensive and time-intensive, is now neither. The tail has shifted, and so has the business case. "If everything is cheaper, build everything" as a 3rd-quarter goal is just bad business, for a number of reasons.

The "first to market" impulse on an AI timetable means even a good idea will be executed without any meaningful ideation. No creative forethought, no use case examination, all "build it and they will come" energy with no backup plan if they don't.

There are no business textbooks that suggest spooning out slop to see if your customers will eat it. Agentic AI has not changed that. If anything, innovating in the wrong direction is as likely to alienate your customers as endear them. Every opportunity space is filling with competitors who are thinking about the customer first rather than ramming out half-sketched feature sets.

Slapping AI whimsically about your product because "everybody is" doesn't make you competitive. If the resulting customer experience is detrimental, they'll let you know with their wallets.

Implementing AI is a choice. Choose wisely.

A helping hand

Agentic AI's real strength isn't in execution. For organizations that provide services either alongside or instead of products, it's in giving the everyman access to sophisticated systems. No more master's degree in toggles-and-configuration required. Empowering the average user who just wants to get stuff done should be your intent.

An AI Agent can consume all the printed documentation on a feature in the time it takes you to read this paragraph. Specifications, live examples, reddit posts and google hits, all in response to the prompt "how do I create a time entry in this thing I've never used before". It can walk you through step-by-step, or perhaps even execute the request itself.

Taken a step further, it can mine reams of data to answer simple or complex questions, then return the results in simple prose or sophisticated charts and images.

AI's true superpower is in reducing the mass of technical debt every modern employee now carries. From "is this floor wax correct for this surface" to "what is the proper input sequence to reset this robotic arm." Sure, you'd expect employees to hold some of this knowledge natively but not everything all at once.

What was once an insurmountable technical hurdle is now a conversation. Agentic AI's customer benefit is in reducing both tedium and complexity at once, in resolving the busywork and overhead quietly in the background, in completing the fifteen tedious steps needed to accomplish one simple thing or many complex things in sequence.

That, my friends, is the problem you should be solving with AI. The "getting the minutiae out of people's way so they can stay in flow and succeed."

Fast is relative

AI is fast, but true innovation and understanding are anchored at the speed of human thought.

AI can generate a thousand different paths towards possible success, and if you don't like any of them, a thousand more. Success isn't in picking something from the list and having an AI build it, it's picking the correct one and implementing it thoughtfully, carefully, with the same amount of consideration you put into pre-Agentic.

Again ... can doesn't rise automatically to should.

Earning our own lessons

Agentic AI took a front seat in developing AbleTime, and it taught us a lot of hard lessons. Everything is faster, but not necessarily easier (or maybe it's that the hard parts are just different). Some AI tools were fantastic. Others proved problematic, feeling half-considered and more prototype than product. They did serve as a cautionary tale, though.

Occasionally the work outpaced our intent. It meant refactoring, cleaning up features that would have arrived all-at-once in the world before. We slowed down, and in doing so our delivery improved. It gave us a chance for perspective to actually see the forest among the trees, instead of just wandering among them.

Slow meant we had time to identify which Agentic features we wanted to develop, which competitor features we could improve upon ... and the many that were just meaningless faff and bother.

For us, the choices were carefully mapped out and user-vetted before anything was built. Agents as timekeepers, tracking work quietly in the background ... but only saving drafts that have to be accepted by the user and invisible everywhere else. Agents acting as office assistants ... but asking permission before committing any action, and customized to the user's personal style.

In my next blog, I'm going to talk about the process we used in AbleTime to pick what we felt were the best use cases, how we implemented them, and why we're confident they will make a huge difference in people's lives.