2025 Predictions

My thoughts on where the ball is headed in tech, startups, and macro

In my last post, I looked back at prominent themes emerging in AI that I think helped shape the narrative around the technology over the course of the past year. I expect 2025 will build off of the tremendous momentum of 2024, and build on those themes in several transformational ways.

Over the last few years, I’ve spent a lot of time researching new technology and speaking to startup founders working to change the status quo. From that data, I’ve developed ten predictions that I think will have the most profound impact on technology and our society next year.

1. AI will act as judge but humans will have a role as jury

   1. The need for AI to grade itself will increase as the demand for testing and evaluation skyrockets in this newly agentic world. We’ll see a number of AI companies enter the evaluation space given the size of the opportunity

   2. Humans will play a key role in this new era. Unable to rely solely on AI accuracy rates (even if checked by other AI), human-in-the-loop approaches will be required to get agents into production for business use cases. The addition of human-led observability and error monitoring will ease the enterprise onramp for agents

2. Legal AI will have its “meta moment”

   • Lawyers will be using AI to fight AI. The technology is exploding across the legal landscape as a copilot to help lawyers conduct case law reviews, review contracts, and handle routine research more efficiently

   • At the same time, tensions are rising and lawsuits are piling up to debate and determine what constitutes fair and ethical use of AI across industries, particularly in areas like entertainment and media, where content is protected by copyright laws.

   • LLMs have already exhausted accessible public datasets, so the race to ingest greater amounts of training data will intensify this year. AI innovation will press on, more data will be consumed. Where exactly that data comes from is sometimes unclear and will remain so, leaving foundation model providers open to arbitration, and fueling the ongoing debate as to whether copyrights have been infringed during the training process. Somewhat ironically, AI legal tooling will likely be applied in prosecuting and defending many of these cases

   • AI, therefore, will serve as a trusted copilot to determine the ethics of its very existence

3. AI pricing models are still fluid

   • We’ll see a spike near-term in volume and outcomes-based pricing for most agent companies, given prevailing questions around accuracy and cost of implementation / ROI tradeoffs. However, once economies of scale kick in, chip production ramps, and cost of compute begins to normalize, regular value-based recurring SaaS pricing models will come back into vogue. I estimate that when the gating issues of cost and trust are solved, the reliability of SaaS revenue streams is too lucrative to abandon

4. Wealth management (and women) will be the key to consumer fintech dominance

   • You’ve heard of new money and old money, but what about young money?

   • More money will be passing from the Boomer generation to the Millennial and Gen Z generations than has ever transferred across generational lines before. An estimated $31 trillion in assets will be handed down, with the majority going to women. That represents a huge opportunity for financial wellness and wealth management offerings catered towards Millennial and Gen Z women. Fintechs will need to adapt to meet this need

5. Labor relations will tighten

   • Labor disputes will be heightened as AI and automation grow in prevalence and employee-company relations become more tense. This will have an impact across industries, but tensions will be highest in unionized sectors where employees are already organized to safeguard their rights

   • There is growing fear that automation will eradicate certain jobs and we are barreling towards a massive labor market dislocation. As we saw with SAG and the longshoreman strikes in 2024, employees are reasonably pushing back on the spread of new technologies that they believe may one day replace them and their livelihoods.

   • I still believe that AI will be a net positive force and a job creator, but also agree that jobs of the future may look different when compared to those available today. Without explicit knowledge of what the future will look like in each industry, it is safe to assume that employee frustrations and concerns will come to a head in 2025, with a greater number of delegations asking for additional protections.

   • This will invariably push federal policymakers to take a stance. While the current administration has implemented a number of technocrats in its ranks and vows to be hospitable to innovation, a large portion of their support base works in industries dabbling with automation. Straddling that line will be difficult, so I predict additional policy measures will be put in place that ease the fears of constituents while simultaneously supporting an agenda of American superiority in AI

   • What’s most interesting to me is the profit-sharing request that surfaced in the longshoreman’s strike. Employees pushed for a new business model that would allow them to share in increased profits from automation. I think we are a ways away from mass acceptance of employee-owned businesses and Universal Basic Income (UBI), but some form of a redistribution of profits for employee retraining and upskilling could be interesting

6. AI has been given agency and its autonomy will increase… for very specific tasks

   • AI agents will continue to be a core focus, and most will tackle a very specific use case rather than serve as general purpose agents. Proving an agent can do an explicit task with a near-perfect degree of accuracy will be a crucial stepping stone to eventually unlock general-purpose horizontal agents in the future. Before agents can be trusted to perform a full spectrum of tasks, they must first prove that they can execute a single task, consistently, without error

   • The end state will eventually include both horizontal (orchestrator) and vertical (including function-specific) agents working in tandem, but use-case specific agents will be the first step

7. Impacts of the new presidential administration will be broad and deep

   • The Trump administration’s regulatory stance on AI will have a positive effect on US-based innovation. By promising to not curtail the pace of development, signaling fewer guardrails, and even spinning up a US AI Task Force, the administration is focused on fostering domestic AI capabilities to best international rivals like China

     • This laissez- faire approach will be a boon to our national superiority agenda, but will drive an ever-greater need for protections. AI trust, and safety and cybersecurity solutions will be more critical than ever

   • Defense tech, space tech, crypto and other emerging categories will experience tailwinds from the administration’s proposed deep tech friendly agenda. These industries (+ the need for GPUs and semiconductors) will drive a greater emphasis on domestic hardware production and may even continue the trend of returning some manufacturing capabilities onshore

8. Demand for talent with AI/ML engineers will continue to exceed supply

   • There simply aren’t enough people with deep knowledge in the field to meet needs. Acqui-hires will pick up and retention of engineers with AI/ML experience will be a key focal point for companies who manage to attract them. Due to these supply constraints, AI/ML talent will continue to command a high price.

   • A knock-on effect of this trend will be a greater proliferation of “workaround solutions” that directly address the dearth of talent:

     • Some companies are responding by building tooling to empower traditional software engineers to build AI products without specific expertise like LastMile AI and Nomos

     • Others, like Uber, took a unique approach and launched a coding gig economy for AI Projects to attract moonlighters and part-time workers to help meet the growing demand for AI model training and data labeling services

9. The lean startup takes on a whole new meaning

   • AI-first teams will stay leaner for longer. With the proliferation of copilots and knowledge worker productivity tools, teams will be able to accomplish more without dramatically scaling headcount

   • This applies directly to fields like software engineering which typically makes up the majority of the first 10 hires made at a technology startup. Startup teams will be scrappier in the early stages and keep human capital costs to a minimum

10. M&A Heyday

    • The pace of M&A will pick up under the new administration and change in leadership at the FTC

    • We will likely see more consolidation in this administrative environment across all stages. From FAANG acquisitions to startups acqui-hiring other top talent, and everything in between.

    • Incumbents that were caught flat footed and haven’t adopted an AI-first strategy will switch from offensive to defensive positions, using acquisitions and in some cases possibly even gating API access to thwart upstart competitors

These are the areas I’m most excited to track in the coming year. Some are clearly supported by preliminary data while others are highly speculative. I’ll be following developments in these categories closely. 2025 is set to be a momentous year for startups and the convergence of these trends could cause a transformational shift in the startup landscape.