20VC with Harry StebbingsAtlassian CEO: Enterprises Won't Trust Chinese AI | Why The Jury’s Still Out on Specialised Models
20VC with Harry Stebbings
Agents and interfaces 0:00
Giving up on the user interface is called a brainless bet. Some tasks will be handled by agents, especially simple ones where you just want the result. But in enterprise work, design and experience matter more, not less. Atlassian is investing heavily in both because the best products will still need strong interfaces, strong context, and ways for agents and people to work together.
Context and design 9:00
A top priority is the enterprise context graph, along with new design work for the agentic era. Text boxes alone are a poor interface for powerful tools. The hard problems are provenance, accountability, and knowing when content came from a person, an agent, or both. The same applies to business processes, where you need to steer exceptions without seeing every detail all the time.
Synthetic content may dominate 15:30
He says AI-made content will likely become normal. You may not be able to tell what was touched by AI and what was not, and he does not think that by itself makes content better or worse. The point will be the quality of the result, not whether a tool helped make it. Human judgment will still matter when something crafted by hand needs to stand out.
Chinese models and trust 17:01
He says some customers push back on what they call Chinese models, but not on open models in general. At Atlassian, they use open models such as Neotron, Mistral, and others, but none of the Chinese models go into production. He treats model choice as a customer problem first and a geopolitical question second, and says enterprises will have a hard time refusing cheaper open models when they know what is in them.
Specialized models under test 21:30
He says the jury is still out on whether building specialized models is worth it. Some narrow cases do well, especially where there is proprietary data or highly specific work like robotics, but many teams are only fine-tuning open models and calling them their own. He thinks general models are improving so fast that many special-purpose efforts will not keep their edge unless they can keep beating the best general model on cost and performance over time.
Routing and cost control 30:00
Large companies will keep doing model routing inside their own systems. Smaller businesses can lean on routers, gateways, and harnesses that make those choices for them. The aim is to blend cost, quality, and speed for each feature, then push the savings back to the customer. Cost matters most when a feature grows from a few users to millions.
Inference and imagination 32:01
Model inference is still expensive, but the price should keep falling as chips, data centers, and distillation improve. Even so, demand will rise as the cost drops. The speaker expects AI to raise product quality and push business toward imagination over information. The hard part now is making time to think deeply in a world that keeps slicing attention thinner.
Design and long races 45:31
Confluence is held up as proof that content alone is not enough. Even though you could write notes in many places, the business has kept growing because of design, experience, purpose, and how well the product fits into other workflows. The same idea is applied to Jira and the teamwork collection. The point is to keep improving every part of the software, because what counts as good design changes over time. Atlassian intends to keep running its own race and stay in it for decades more.
Energy and founder life 49:00
Founders are described as taking on a hard job that can strain money, reputation, and mental health. The speaker says the work only lasts if you actually enjoy making products or services, and if you know where your energy comes from. He says he is not lonely because he works with smart people and still sees Scott regularly. He also says the people around him are a major reason he keeps going, and that many people who enter the field for the wrong reasons cannot sustain the effort.
Money and self respect 54:31
He says the future price of inference will likely fall farther than expected. Money is described as both a wonderful thing and a burden, because it carries a responsibility to do useful things for society. When asked about raising children with money, he says he does not know the full answer, but he thinks humility, curiosity, gratitude, and growing into your role matter. He also says to give yourself peace, because tech people often set the bar too high and turn that into a new kind of stress.
Slow thinking ahead 57:30
Asked what he still needs to learn, he says it is how to make room for slow thinking in a world that is speeding up. Phones, social media, and constant notifications all pull at attention, and he thinks the challenge will only grow. The task is to slow down enough for deep thought while still moving fast in the real world. He closes by agreeing that a good person helps without looking for an immediate return.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
