1. The future of agentic coding is a) multiplayer and b) more than a chatbox. I'm excited about a future where multiple humans can collaborate with agents in a single session. The UX for working with AI-generated artefacts also needs to improve. For example, editing or commenting inline directly on the plans or code.
2. Humans and AI should be living in the same tools for tracking work. It doesn't make sense to have one orchestration layer for tasks delegated to agents and another place where human tasks are captured. Nor does it make sense for tickets in project management software to be picked up by humans who then work on them privately with agents on their local machines. There should just be 'work that needs to be done' which can easily be passed between humans and agents. I think Linear is well positioned to be the de facto solution for this.
3. Until product development processes get redesigned around AI, there won't be a measurable increase in output. Right now, the focus is on using AI to improve individual productivity, but that does not directly translate into team velocity (except for greenfield projects or solo devs).
4. Engineering will divide into two tracks: product builders and platform/infra/safety. Product builders will be expected to be part engineer, part designer, part product manager. The platform track will be responsible for the foundational layer and guardrails that let everyone ship to production safely and at high velocity.
5. Companies that push "everyone should build" but underinvest in platform/infra/safety will struggle to retain engineers as they become saddled with the unsatisfying job of cleaning up everyone else's AI-generated mess.
6. AI providers will build services that establish a moat around agentic development. Today, there is little friction in switching between models or harnesses. I expect providers will begin hosting apps developed on their platforms, with managed access controls, because that creates switching friction.
7. We'll see greater adoption of local or cheaper models for coding as token spend becomes unsustainable and as subsidised plans disappear. Right now, most people default to the newest and smartest model available. Over time, things will settle into more of a balance between cost and capability.
8. There will be a catastrophic data breach of a popular AI tool leaking sensitive internal company information such as emails, meeting notes, or docs. This will push companies to adopt local models and restrict use of 3rd party tools.
9. Unfortunately, we're likely to see more layoffs at big tech companies, not because AI is replacing humans directly, but because companies need to offset rising token spend without a clear revenue boost.
10. We find a way to reduce AI-generated slop content. This is more of a hope than a prediction. Even if no technical solution emerges for spam bots, hopefully people realise that posting AI slop hurts their personal brand rather than building it.