Do you remember the first question you ever asked ChatGPT?
Mine was about music recommendations. It did okay.
I asked it the same question tonight and it told me one of my favourite bands released a new song a few days ago. I love that band so much (American Football) but didn't know they had new music. I immediately searched and pressed play.
Three years ago, ChatGPT acted like a music critic, telling me about bands people with my stated taste in music tended to like. This year, it scoped the problem, then searched the internet to find a new American Football song of all bands, absolutely nailing the brief. How'd it know to do that?
The Spaceship Universe and Spaceship Earth portfolios invest Where the World is Going (WWG). You know the drill: businesses aligned with long-term structural themes, with attractive growth characteristics and leadership teams that think and act like long-term owners.
Recently, one of these long-term trends has impacted many of the others, and you can guess what it is. AI's been doing for software what it did for my music search: evolving to act near-autonomously, and spooking the market.
What is an AI agent?An AI agent is a software program that can act on your behalf. AI agents are generally powered by large language models (LLMs), which are AI systems trained to understand and generate text, including software code.
AI agents act as self-directed software. They're different to simple chatbots that are built on pre-defined systems and pattern matching, such as Clippy or OG Ask Jeeves. They're also different to predictive engines, such as the 'people also bought…' and 'people like you may…' features in software and websites you might have interacted with.
Instead, an AI agent can autonomously decide its next step, plan it, and then take it, with little human intervention.
For example, asking ChatGPT for a list of good holiday destinations will see it return an answer based on its training. That's generative AI. You ask the question, it answers, and considers the case closed.
But asking ChatGPT to recommend a holiday in your price range, based on your unique interests and timeline, and then it deciding to search the internet, shortlist options, compare prices and tailor recommendations based on what it knows about you - that's agentic AI. You ask the question, it takes a multi-step autonomous approach to execution, breaking the goal into subtasks, using tools, making decisions along the way, and iterating toward an outcome.
Or, put another way, AI can help Netflix's algorithm show you what to watch next based on what people like you already liked; while agentic AI can theoretically traipse through the internet and return a personalised watchlist complete with streaming services, running times, and matching wines to pair it with (okay, sign me up for that.)
TL:DR; What is agentic AI?So agentic AI is when AI can plan and take a series of different actions to achieve a goal, generally without much human input.
Who creates AI agents?In general, anyone with technical capability and access to AI models can create an AI agent. That's where it gets a little bit hairy.
Until now, software has generally been created by human coders with specialist knowledge, supported by teams behind them. You needed someone with an idea, someone with funding, someone with know-how, and someone with marketing skills to make sure your software made it to production and found customers.
Now you can arguably go straight from AI model to customer via agentic AI, and cut out most of the middle people. That's one of the reasons why markets have been pretty choppy recently - because software companies are some of the biggest on the stock market in recent years, and they suddenly have competition that not many people saw coming.
What is SaaS?To cut a long story short, SaaS stands for Software-as-a-Service. Pre-internet, software companies delivered their products via floppy disk and CD-ROMs - that is, on physical devices.
Then, when the internet became mainstream, it became more possible for businesses to download or sign up to software programs straight from websites. Then 'the cloud' became ubiquitous. This meant that software companies could sell their services, running them remotely to virtually anywhere. Typically, they'd offer access with subscription pricing, and the whole model became known as SaaS.
What are some SaaS companies?SaaS companies have become some of the biggest, and near inescapable in modern life. You might use Microsoft if you have an office job, or Canva if you're a creator. Shopify is one of the homes of online retail you might have shopped with, while Salesforce is where many of the corporate companies such as banks and airlines keep track of customer information, which may include yours.
All of these companies sell Software as a Service, enabled by the cloud.
Who can use AI agents?AI agents can be used by basically anyone with a screen and an internet connection - business or customer.
Recently, Anthropic released Claude Code, its agentic coding tool that lets AI write and ship software autonomously. You give it a task, and it runs with it. So developers can now build and scale at new speeds, with less of a need for companies whose bread and butter has been doing the same things, but slower. These companies also tend to charge per seat, which means if fewer people need them because they can DIY with Claude, then future revenue can look a bit shaky.
Cue the SaaS-pocalypse.
What's the SaaS-pocalypse? Let's ask Jason Sedawie, the Spaceship VP of Investments."Yeah, it's causing an event that investors are calling the SaaS-pocalypse — the software-as-a-service implosion. We've seen stocks in the software index down around 30% in the last couple of months, basically because of Claude's AI agent.
Investors are extrapolating: well, if AI agents can do this on your behalf without using software, what's software worth?
For us, we've always concentrated on companies with unique data, because an AI model is not really worth anything unless you've got unique data. Companies in cybersecurity like CrowdStrike have interesting cybersecurity data, and a company like FICO, which does credit lending standards, has that embedded data as well.
We think there are some risks in software — maybe the prices they can charge going forward might not happen — but we think they become more efficient as they use AI tools themselves.
So while the market has kind of overreacted in this case, we believe those companies with unique data are still well positioned to do well in the AI world going forward."
We asked Jason to tell us more.
"The bear case on SaaS conflates the code with the product. Yes, AI makes code cheaper to produce. But that misses the point entirely. What enterprise software companies actually sell is proprietary data, embedded workflows, compliance, and trust. The accountability of having someone to call when things go wrong. Open source has been free for decades. Enterprises still pay for licensed software. That tells you everything you need to know about what they're really buying.
AI agents will drive greater software usage, increasing activity and their relevance.
The more interesting question is what AI does for software companies, not to them. The best platforms are already deploying AI natively, agents are becoming the product, not a feature. And when agents run continuously instead of humans logging in intermittently, the volume of activity flowing through these platforms multiplies. More activity means more data, more complexity, more risk. Risk demands governance and guardrails. Governance and guardrails require a trusted vendor with accountability. The companies already embedded in enterprise-critical systems are the natural home for all of that.
Network effects also compound in an agentic world, not shrink. Proprietary data, deep integrations, compliance certifications, these took years to build and can't be replicated by a model trained on public data. As automated decision-making carries higher stakes, trust becomes worth more, not less.
We're confident software is still Where the World is Going."
...but as long-term investors we know to watch for the signal in the noise. Interested in what else is in our portfolio? Check out the Spaceship Voyager portfolios and see if they align with where you want your world to go.
Important: We're sharing our thoughts on companies in which Spaceship Voyager invests for informational purposes only. As at 2 March 2026, some Spaceship Voyager portfolios invest in Microsoft, Shopify, Salesforce, CrowdStrike and FICO. We think it's important (and interesting!) to keep you informed about what's happening with Voyager's investments. However, this information is not a recommendation to buy, sell, or hold any specific company or security. Past performance is not a reliable indicator or guarantee of future performance.
The information in this article is prepared by Spaceship Capital Limited (ABN 67 621 011 649, AFSL 501605). It is general in nature as it has been prepared without taking account of your objectives, financial situation or needs.
