Years of SaaS software have messed with people’s heads when thinking about AI. Companies & schools are used to outsourcing R&D - software vendors build products & organizations implement them. For AI, no one has special insight. Folks will need to do R&D themselves to get ahead. It is crazy to look at a 18 month old General Purpose Technology and say “well, if no one has a fully implemented publicly available product demonstrating 30% cost savings or [insert KPI] then AI must be a bust.” I promise you some companies are figuring this out, but it requires an actual innovation effort snd trial and error.
I strongly agree, but the flip side to this is that no matter how good the models are, I think something roughly like probably 30 to 50 percent of people who could benefit from the simply have no interest in even opening the chat GPT app, let alone experimenting enough that it becomes deeply integrated into their workflows. Whether it's temperament, tech aversion, or the cognitive tax of each additional click they have to do to get data in and out, I think this will basically remain true. This means that the real product opportunities are the ones where you take what the power users learn via user-led innovation and encode that into a completely streamlined, nearly invisible product experience for which enterprises are willing to pay a nice premium. That's how I'm thinking about it at the moment.
If you can’t demonstrate even a feasible way to 1) make any kind of objective difference while 2) turning a profit yourself after 3) spending literal trillions of dollars in a technology, maybe it doesn’t matter that it’s been 18 months? Because the reality is that none of it is, in any way, new tech. It’s being used at scales that are higher than before but we’ve been using the same fundamental models and techniques for at least a decade now. Trying to argue that “hey we’re still figuring this out” isn’t exactly a fair take. We know, pretty damn closely, what this tech is capable of. That’s why I’ve been so critical of some of the use cases: I know pretty deeply it’s not really capable of doing it. What we’re “figuring out” is if people can be fooled, not whether the tech can be improved. It can’t, because it’s up against some pretty fundamental limits. And we’re learning that “fast marketing” isn’t selling this tech, to the surprise of nearly zero people, except those who have a vested interest in the tech.
The past 30 years of IT tech has been dominated by traditional software engineering, where 'experiments' mean A/B test the customer experience. And even that is hard. Most companies, even the innovative ones, simply do not have the leadership experience and the internal culture of experimental science to manage and develop AI technology. The situation resembles that late 90s, where companies struggled to adapt to the new browser technology, while startups like Amazon were pioneering the future we live in today.
That's spot-on. I come from the biotech world: AI isn't a technology, it's an innovation platform. More akin to discovering a new metabolic pathway than to the identification of a new active principle. There is work to be done to create tools, pipelines, integrations, etc. There is a need to rethink processes and systems surrounding processes that are touched by AI. There are new failure modes to take into account. I personally rolled out LLMs in a startup I co-founded. It is an absolutely transformative platform, but it's a far cry from the plug-and-play feeling of AWS Lambda or Docker.
100% agree and I see it as a significant risk to organizations that might slow AI efforts. I think there's a number of factors that are driving that sentiment. The rate of development and user adoption has been so much faster than any previous general purpose technology. The marketing machines have been on steroids using fear to drive sales. I spend every day thinking about this space and how to support AI adoption and I still feel like it's difficult to stay in the loop. Leaders outside the space are exhausted by the pace and hype. When you combine those factors, it's easy to believe if there was value, it would be there already and it's not worth keeping up with the exhausting pace of change beyond the day job... wake me up when you have something real.
With LLMs being as accessible and versatile as they are now, I am a bit hesitant when it comes to investing time and money in SaaS solutions for something I can directly execute within a frontier model environment. Granted, I have been spending a lot of time with ChatGPT-4, Sonnet 3.5 and Gemini but after observing my own behaviour, I’m curious how that plays out in the months/years to come for organizations. Will the question be “Can I do this directly using the LLM or does my solution require an in-between vendor solution?” I sadly feel like this won’t be good for a wide variety of SaaS companies. Hope I am wrong!
I’d look at how strategic planning is conducted inside companies. There’s usually more projects in the backlog than there’s budget to realize all of them. Between carving some budget right now for something highly uncertain and waiting to see what others come up with to de risk the investment (while investing in projects with more certainty around their ROI), it is probably tempting to wait, especially if your performance is scrutinized quarter to quarter. Ideally there’s an innovation budget for experiments, but the cost of R&D around figuring out the proper LLM use cases and deploying them to production (and maintening them) might be a barrier. We all know that many AI projects never make it past the POC stage.
100% agree on the fallacy of expecting AI will be like SaaS. But disagree that for AI no one has special insight. There are a good number of companies that were doing AI R&D prior to LLMs and still ahead of the curve. It requires deep-enough knowledge of AI to spot them.
You're right - the SaaS mindset doesn't fully apply to AI implementation. While off-the-shelf solutions work for many software needs, AI requires a more hands-on, experimental approach. Organizations can't just wait for a perfect, pre-packaged AI solution to emerge. Instead, companies need to invest in their own R&D, tailoring AI to their specific challenges and opportunities. It's premature to judge AI's potential based on readily available products at this early stage. The real innovations are happening behind closed doors, through trial and error. Success with AI demands internal expertise, creative problem-solving, and a willingness to iterate. It's not about finding a one-size-fits-all solution, but about developing a unique AI strategy that fits your organization's needs and goals.
Future of Work Speaker | Payroll & HR Tech Expert | Pay Transparency | Author | Linkedin Top Voice
1moI’m not so sure. Why do companies buy software or saas? Because it’s not their core process, and they don’t have the knowledge nor the R&D capacity/resources/money. I’m in HR: no company builds their own HR & payroll software. AI doesn’t change that - they will look at their (cloud) vendor. While they will do it for their primary process. We’re vastly overestimating how “easy” or “accessible” these technologies are for the majority of users outside of the tech/Linkedin bubble.