In every industry, people are rushing to roll out new AI apps or add AI capabilities to existing apps. And you can’t blame them — the results speak for themselves.
In a recent survey, we found that 92% of IT decision makers believe that deploying these technologies will lead to meaningful business outcomes within the next 12-18 months.
We at SnapLogic agree, and so do the customers we have already helped to deploy AI in production.
However, it is sometimes worth reminding ourselves that this current wave of interest in AI is still brand new. The launch of ChatGPT to the public was arguably the trigger event — but that was only in November 2022! Since then the market has evolved very rapidly. OpenAI has continued to release new models, as have all sorts of other providers — too many to name, in fact, with new ones emerging all the time, most notably the recent arrival of DeepSeek.
Moving fast on shifting ground
This rapid evolution is hugely beneficial to AI customers, of course, as we all get to take advantage of the new capabilities, raised limits, and multi-modal features of the models. However, fast-moving environments like this one also present a challenge to anyone trying to build services that rely on or integrate AI functionality, as the foundations are constantly shifting.
For an example of how things can go wrong, look no further than buzzy AI startup Humane. Their Ai Pin was an intriguing device, promising to integrate AI-powered capabilities seamlessly into its users’ lives. At one point, Humane was reportedly valued at $850M. Unfortunately, seven months in, Humane has been forced to sell itself to HP, leaving users with just ten days to back everything up before their devices stop working.
Outcomes like these are sad to see in the consumer space, but the Humane Ai Pin was after all pitched as a gadget, a nice-to-have, not a must-have. Its demise leaves its users out of pocket and presumably disappointed, but no worse off beyond that. In many domains, such an outcome would have far worse consequences — think of an AI-enabled medical device, or an AI-powered HR workflow within a company. Especially in the enterprise space, there may well be contracts and obligations involved, including penalties for non-compliance.
This combination of rapid obsolescence and long-term responsibility is why it is very important not to forget about the future in the rush to design for our exciting present moment. When we talk about “future-proofing” a design, we mean considering carefully what likely future changes may occur over the product’s lifetime. Products start to become obsolete as soon as they are released. My own car is not that old, but its onboard power and data ports are all USB-A, while my phone is USB-C. Luckily A to C cables are readily available, but newer cars ship with USB-C — which due to legal requirements in the EU, is liable to be around for the foreseeable lifetime of the cars.
Don’t get caught in a dead end
In the same way, it is important to design your new AI service to minimise the risks of it being made obsolete by future evolutions in the market. Some evolutions are predictable, others are not, so it is worth building flexibility into the very foundations of your design to ensure that your service continues to provide value to users well into the future.
For instance, you may have committed to one particular vendor for your project, but later you find out that this was not the right partner. Maybe their strategy does not align with your own, or maybe they even terminate the product line that you were relying on. Or maybe the vendor itself goes bankrupt or gets acquired.
Maybe you committed too tightly to a particular technology or architecture, but later you find out that it’s a dead-end, either because it becomes entirely obsolete, or simply because it does not advance fast enough, or as fast as other alternatives.
Finally, maybe you decided to build within one particular ecosystem, but later discover that what you built is not generally applicable or easy to generalise. If you find that you need even very similar capabilities in an adjacent domain, you need to start your development from scratch with a different toolchain.
Let your partners do their part
Unless you’re an AI researcher, it is of course not feasible to keep abreast of the breakneck pace of development in this ecosystem. Few non-specialists had DeepSeek on their radar before this year, for example. This is why it’s worth working with partners that help you future-proof what you build, in particular by partnering in turn with as many different vendors and technologies as possible to preserve your flexibility to evolve in the future. SnapLogic works with Amazon Bedrock, which in turn integrates with a wide selection of the most popular LLMs. We also use it to power our own SnapGPT copilot! The benefit of working with Bedrock is that any new model that is added is instantly available in SnapLogic. When AWS announced the new Nova models during the re:Invent keynote, by the time I had walked back from the keynote hall to the SnapLogic booth, a colleague had already spun up a SnapLogic pipeline that integrated Nova!
It’s important to future-proof your AI investment, but it doesn’t have to be a drag on your productivity or your speed to release new services. If you build with the right partner, you can avoid getting locked in as AI continues to evolve rapidly, thanks to SnapLogic’s LLM-agnostic architecture. We will help you — and more importantly, your users — take advantage of the best and latest generative models, with full compatibility across your existing and future technology stack.
To find out more about how SnapLogic AgentCreator can help you deliver your AI-powered service fast AND make it future-proof, pre-register for Agentfest on April 17th, SnapLogic’s virtual summit dedicated to exploring the transformative power of AI agents.
