From Copilots to Autonomy: Jeremy Soo Maps AI’s Evolution

Jeremy Soo unpacks AI's next wave: autonomous agents, data infrastructure barriers, and talent gaps. Learn how businesses can ethically scale invisible AI

The most transformative AI advancements will fail without human systems to harness them, making organizational "conveners" the unsung heroes of enterprise AI adoption.

Jeremy Soo, CTO of Nex AI and strategic leader in AI infrastructure, embodies this tension between technological potential and human implementation. His insights reveal that breakthroughs like autonomous agents and data ecosystems are meaningless without bridging talent gaps and fostering ethical ownership across organizations.

In this interview, Soo dissects near-future disruptions, from invisible AI systems to revenue-generating agent swarms, while pinpointing why culture, not code, dictates scalability.

What do you see as the most transformative AI technology in the next 2–3 years, and how can businesses prepare for its impact?

In the near term (2-3 years), I see Autonomous Agents being the most transformative for many businesses. In the early days of AI, the best tools were simply co-pilots, providing suggestions and answers when prompted. Today, a decent chunk of the tech-forward market has begun to adopt AI Agents, tools that at their core, put an AI model in charge of deciding how to respond to a given circumstance or prompt.

Soon, there will be an extrapolation of this technology, with more and more functions being outsourced to fully autonomous AI systems that decide on objectives, optimise from their own research/collected data, and refine with minimal human intervention. Essentially invisible AI that runs in the background 24/7, which is already happening for customer support, essential alert roles, etc.

As AI adoption accelerates, what are the biggest barriers to scaling AI solutions across an organization or industry?

The biggest barriers to scaling AI solutions is talent. Namely, expertise at the in-house level. In many enterprises and industries that would obviously benefit from AI, e.g. retail, manufacturing, logistics, etc, that are traditionally more old school, internal tech teams are virtually non-existent.

It is more common to outsource and maintain a 1-2 person duo to fix “IT problems” as they go. This “translation layer” between end-AI users, and the people who build it leaves a lot of room for “hobbling”, messing up obvious scaffolding implementations, tooling development, etc.

At the same time, the very best researchers and engineers at the frontier labs are constantly pushing the boundaries of what can be done with AI, yet little of it gets past a Proof of Concept stage in most large enterprises.

How can companies strike a balance between rapid innovation in AI and managing ethical or regulatory risks?

Companies need to take a holistic approach to AI and adopt it throughout the company across all functions to 1) foster collective ownership of adoption, and 2) get skin in the game if anything goes wrong.

The current biggest problem in many medium and large enterprises that I have talked to and/or worked with, is that one or two proactive individuals have to feel like they are sticking their neck out to try a new technology, while most others can sit by and blame them if anything goes wrong.

In companies where leadership is strong, and conviction trumps political fear, we see much faster AI adoption led first by a top-down approach, and then individual guardrails start to kick in because people have to find ethical and compliant ways to use the tools they are expected to use.

Of course, the baseline is having a company wide policy about what is allowed and not allowed, dependent on industry. A news company definitely should not be generating fake images, while a manufacturing company using GenAI to produce high quality 3D mock-ups is expected and certainly smart.

What role does data infrastructure play in enabling or limiting the growth of AI-driven technologies?

Data infrastructure is the most important and also confusing part of adopting any AI technologies. From my experience as CTO at Nex AI, this is the biggest bottleneck in companies trying to adopt AI. Data infrastructure can be more simply understood as “Where data is”, “What it looks like”, and “How it gets processed”. In most companies, barely one of these 3 questions can be answered with certainty.

This is usually partly due to a process issue, where people themselves are not clear how data should move around the company, and so codified systems end up mimicking the convoluted data architecture of real life. In truth, having a robust data infrastructure, with clear isolated layers, e.g. a data lake, some transport layer, a high-visibility transformation (processing) pipeline, the right database solutions, etc. is 90% of the work in helping any organisation adopt AI.

Turning sparse data into clean structured analysis-ready datasets for enterprise systems (e.g. ERPs, CRMs, custom software) is the most important part of every transformation project we have taken on.

In what ways can AI be used not just for efficiency, but to drive new revenue streams or business models?

The most obvious way to use AI to drive greater wins, and not just smaller losses, is to use AI to do the same things humans would do to get more business, just at much greater scale than ever before. This translates to AI SDRs, AI BD efforts, AI-powered marketing, AI personas as digital influencers, etc.

Of course, selling AI tools itself, coupled with vertical non-AI infrastructure that a company has built out in the pre-AI era, is a great way to easily jump on the train, make more revenue, and get data to refine their tools. This is what Vercel has done with v0, for example, or what Intercom has done with their pivot.

I think another key point to note as we move towards more AGI-like behaviours, first, agent swarms, then autonomous agents, etc, new revenue streams will naturally arise due to the cost of labor dropping dramatically. It is fairly straightforward to say that “tedious” businesses that are OpEx heavy, will become more and more attractive as time goes on.

I think this ties into the next wave of service-based businesses. Product businesses have a very obvious next step which is to couple complementary services powered largely by AI with their core offerings.