For most of software engineering history, the vast majority of applications were built using deterministic approaches. We even aimed to make our systems idempotent, ensuring that the expected state continued to be.
AI was used, but for specific use cases… predictions. We acknowledged that it would be imprecise and that the answer was an estimate, not a fact. We understand we may not get the same result with subsequent runs. We even used this to see a range of potential outcomes with Monty Carlo simulations.
But, AI wasn’t used to replace basic workflows and especially not things that required consistent, replicable results, like transactions. That was, until OpenAI released ChatGPT and suddenly “AI” became a household term. Many startups jumped on the bandwagon, ensuring they had a .ai website, in a almost bizarre rehash of the old .com boom in the 90s.
Companies started to invest in using AI for workflows that weren’t really suited to it. But they didn’t use AI built in-house on their own data, rather they used a CloudAI, that was trained on the entire internet & they paid a subscription for access.
The argument was simple. The subscriptions were super cheap, we could spin up functionality faster than having our own developers build something. But behind that was also the tension most CEOs felt with their boards seemingly asking always, “what is your AI strategy?”.
FOMO was driving their impetus to integrate cloud AI into their business. The prevailing mindset was that it would be cheaper than the human processes they had in place already. So what if it hallucinated once in a while and promised our customers deals that were unprofitable? We will save on labor costs!
But, companies found that most of the time, the ROI on these cloud AI efforts wasn’t there according to an MIT study. The promised savings and better outcomes never arrived. But at least the subscriptions were cheap, so let’s continue to push and hope that it will work out in the end… because volume!
At the same time, CloudAI providers like OpenAI and Anthropic have been raising billions of dollars from investors, with the promise that they too would have a ROI. But, the funding is just keeping the lights on. OpenAI is losing double-digit billions of dollars per quarter!
Eventually, investors will insist they won’t continue to throw money away and want to see a move toward profitability and thus an ROI. OpenAI is now trying to run ads inside your chat window. If you’ve been using Agentic AI with Claude Code on your subscription, well, that’s gone too.
Companies are now facing situations where they thought they had a <$200/month subscription that was all you can eat and now they find they are paying $10K/month in API token costs! Developers who might have used AI to make simple tweaks to a code base through all the agents they built now find that inserting one line could cost as much as their effective daily salary.
AgenticAI only makes sense for companies when it is a flat fee and unlimited. Paying per API call or having your prompts throttled completely changes the value proposition of platforms companies have built on top of CloudAI.
We are moving into the reality phase of the CloudAI technology trajectory. While rerunning a prompt because of bad output might have been fine when we had an all-you-can-eat subscription, it looks very different when you pay for each prompt, whether it gave a useful output or not.
It’s likely ChatGPT will follow ClaudeCode’s lead here and move away from the subscription pricing since they lose so much money on them. The problem is the bill is coming due for the companies that had been enticed by the promise of insanely cheap processing. They are now facing the reality of a platform that already didn’t provide an ROI with subscription pricing now facing up to 50X more cost with no improvement in efficacy.
That of course is just individual companies that are trying to use the AI for their own business uses. Companies which were built up to be parasites around the CloudAI providers have their entire cost basis of their business decimated.
This ticking time bomb is approaching for companies that have relied on cheap CloudAI. It’s not that AI in general is the issue… It’s the promise of generalized AI that we can use in our business case. This was never going to be effective, let alone provide an ROI.
LocalAI has always been the answer. From a security perspective, from a cost perspective, and from a data perspective you want it to be in your environment. Avoid the dilution of insight, the high cost, and the protection of your critical data by building your own AI platform. Reach out if you need help.