Agents overlooked, scaling limits overblown: Key AI trends from NeurIPS 2024
Vertex Ventures US
We partner with and invest in early-stage founders building infrastructure and SaaS companies.
Welcome back to Vertex Angles, the weekly newsletter from Vertex Ventures US . We’re a boutique venture capital firm, investing in exciting companies across software infrastructure, developer tools, data, security, and vertical SaaS. If you prefer e-mail, you can subscribe here to get Vertex Angles in your inbox every week.
Programming note: This is our last newsletter of the year, as the Vertex Ventures US team prepares to wind down for the holiday break. Thanks as always for reading, and we’ll see you again in 2025. May your holidays be joyous and bright, and Happy New Year!
This week, Vertex investor Simon Tiu shares his takeaways from NeurIPS 2024, held in Vancouver, Canada last week.
The halls of Vancouver's Convention Centre buzzed last week with a different energy than years past (and not just because Taylor Swift was playing her final shows of the Eras Tour across town #Swiftie). At NeurIPS 2024, in contrast to prior years, gone were the speculative conversations about AI's potential. Instead, I witnessed something far more exciting: thousands of researchers and founders rolling up their sleeves to tackle the nitty-gritty of making AI work at scale.
As an early-stage investor, I see this shift from theory to practice as a sign of a maturing market ripe with investment opportunities for founders ready to build for a future where knowledge can be generated as freely as air.
Despite what “scaling law” doomers may say, the startup opportunities in AI haven't diminished—they've matured, and we’re still in the very earliest innings. The winners of 2025 won't be those with the largest models but those who can build sustainable, defensible businesses on top of AI infrastructure.
As we look ahead, I'm particularly excited about founders who combine technical excellence with pragmatic business execution.
Here are some of my thematic takeaways from NeurIPS 2024:
Agentic Workflows Are Here – Now Let’s Get Practical
One of the most striking aspects of this year’s conference was the sense that an AI-driven future is no longer a question of if, but when. Conversations rarely revolved around whether AI “could” transform industries. Instead, they focused on timeline, execution, and overcoming practical hurdles. In one workshop on “Scaling Autonomous Agents,” researchers and founders argued less about the feasibility of agents navigating complex tasks and more about reducing latency, improving reliability, and ensuring these agents integrate seamlessly into existing workflows. It felt like we’d collectively moved past the conceptual stage; now, the community is determined to iron out the details.
The End of Pure Scale Advantages
If 2023 was about racing to build bigger models, 2024 is about building smarter ones. Over multiple sessions, practitioners from leading AI companies shared how they achieve better results through targeted model optimization and distillation than brute-force training. In general, network effects and benefits occur at the software layer. As in the past, success at the application layer is about PMF power, not GPU power.
Everything is CRUD
While it may sound obvious now, the next wave of successful AI companies won't win through model architecture alone. At their core, business applications are CRUD databases wrapped in business logic - and this entire layer is about to be transformed by AI agents. These agents will prove most valuable where they can orchestrate updates across multiple databases, effectively becoming the business logic itself. The implications are profound: imagine CRMs reduced to agents managing workflows through just email and chat or entire back-office operations reimagined from the ground up. Backend systems as we know them will be fundamentally reshaped. The writing is on the wall, and we're headed for a radically different enterprise software landscape.
Cross-Domain Innovation as a Growth Driver
The most promising conversations happened at the intersections of disciplines. Supply chain experts paired with reinforcement learning specialists. Roboticists collaborated with neuroscientists. These combinations are producing novel solutions that single-domain approaches miss. For founders, this suggests that building diverse technical teams isn't just about equity; it's about capturing unique market opportunities.
The AI revolution is no longer a question of if, or even when. It's here. The question now is: who will build the enduring companies that capture its value?
News from the Vertex US portfolio:
Vertex portfolio job of the week: Digital Transformation Engineer at Tulip
Tulip Interfaces , the leader in frontline operations, seeks an experienced engineer to help modernize the manufacturing industry with its no-code, cloud-native platform for building connected apps. This role is based in Munich, Germany and requires fluency in the German language.
Find more jobs at Tulip here. For more startup jobs from across the Vertex Ventures US portfolio, check out our jobs portal.