To call me skeptical of “AI” would be a compliment. Better yet, just label me as fed up. For the last three years this AI boom has been dominating markets from tech to medicine to art and beyond. The majority of of the hype isn’t warranted. It’s a repeating hype cycle stock pump.

There’s no AGI, and probably won’t be for anyone alive today, if ever. The valuations are staggeringly absurd. The “promise” is always just around the corner, Tesla style. But, OK, somehow it’s all real and business is happening on the back of LLMs. What to do?

About five years ago was probably the high point of IT outsourcing. Back then IT service companies settled into a good balance between knowledgeable people and smart automation to provide high-quality services. There were mature contractual controls around training, quality, auditing, and all the necessary governance required to ensure the customer was getting the services they bought. Nothing is perfect, but that market was pretty dialed in.

Enter AI. Sorry, my bad. Enter scam productivity. I focus on productivity because that’s what “AI In Business” is really meant to deliver. Juice productivity, juice profits. That’s the game.

We know system automation already exists. Skilled programmers already exist. Knowledgeable product managers who balance their risk/reward bets are a cornerstone in industry for decades. People can do the work using all the knowledge and skills they have, and that they regularly continue to develop. But this takes effort and time which means productivity can falter if these processes are not managed well.

AI is here for one reason and that is to increase productivity, and on paper it does a chef’s kiss job at it – just poorly, unstably, with deferred costs, and no guarantee of consistency. It is with those problems service outsourcing completely breaks down. Building processes that rely on AI for a customer-facing deliverable is asking for trouble and will deliver worse results over time than not relying on AI.

Large IT firms spent millions of FTE years designing processes that are repeatable; that can be relied upon to deliver the same results within a very slight variance. No longer. Ask some chatbot a version of the same question over and over, and it will deliver a different answer over and over. Don’t have your top P, frequency, or presence tuned just right? Then hang on for some epic hallucination adventures. Businesses cannot withstand that level of uncertainty.

Chasing AI productivity on the provider side is not like former investments in automation. Automation is by definition repeatable; human designed with strict parameters. Imagine a shop full of craftspeople hand forming aluminum blocks for a customer. Each one would be different and probably useless in the final product. But if the supplier uses CNC automation, they can guarantee exact sameness of the deliverable. The productivity gain here is the lack of defects (and production volume sure). AI fails at this miserably. People need to be the integral part of a service system to guide and produce reliability.

I am not anti-AI. LLMs have a place in business. The technology can add value. I use OpenAI’s Playground to help me brainstorm ideas (not this one) or to play the other side of an argument to provide me with different perspectives. I used AI to help me design and code an iOS app, and also to bulk convert GPS coordinates to physical addresses using Google’s mapping API. Pairing knowledge and AI can produce good results, but not a quality final result using AI alone.

Here are a few points that can help service buyers:

• Don’t accept the output from an AI/LLM as a final deliverable
• If your vendor uses AI in their process, make them prove its repeatability
• When assigning risks to processes, specifically target AI use as a risk multiplier
• Assigning financial penalties for poor performance for using AI won’t be effective, be clear about who is producing your service result upfront
• Seek productively gains outside your critical path
• In data security the principle of least privilege is good policy. Adopt a similar attitude towards AI as a component of your service contract. Only use what really makes sense.

Go forward with AI on your schedule, not the hype cycle schedule. Be skeptical now, so you can avoid being fed up later.

Lastly, once AI has eaten whatever productivity gains it has discovered, the next step will be the principle product itself. But maybe more on that later…