We’ve scoped hundreds of applied AI projects at Fractional AI, seeing where companies are actually investing in AI and partnering with them to move from AI in theory to AI in production.
As we gear up for 2025, we polled our team for their AI predictions:
On investing in AI -
- AI transformation investing will go mainstream in private equity. In the first half of the year you’ll hear about specific deals, and by the end of the year a bunch of firms will be racing to announce new funds aggressively pursuing this strategy. (Chris)
- We'll see a wave of M&A combining early-stage startups that use AI to automate a process with incumbents that have distribution. (Travis)
- Deregulation will increase venture funding liquidity but amplify pressure for higher returns. (Joshua)
- We'll start to see experimentation around the standard investment models as VC investors get nervous on defensibility for vertical AI startups and PE investors get nervous on recruiting technical talent. (Travis)
- We’ll see lean teams and solopreneurs building AI-first ventures, using AI to streamline operations in areas like customer success, finance, and sales, achieving comparable revenue with much fewer resources. (Annie)
- We'll see a divergence in fortunes between AI (which I believe is real) and other investments that I believe are currently in a hype cycle (most crypto assets, meme stocks, etc.). (Travis)
On AI applications and technologies -
- Agentic workflows (where the LLM has some autonomy around the way to execute a task) are here to stay and will rise in popularity, but anthropomorphized agents will be a fad that fades away by the end of 2025. (Eddie)
- Desktop interactions will get to a high degree of proficiency-- OpenAI and Anthropic will have native mac/windows apps that you'll be able to give mouse/keyboard/screen access to and have them do almost anything on your computer; handle your email backlog, book travel, navigate your insurance carrier's website and directories of doctor review sites to find a specialist covered by your plan. (Hugh)
- We’ll all encounter AI voice agents in the wild on a regular basis. I’m not a believer in voice as the new dominant interface for computers - it’s goofy to talk to AI at your desk, especially in an open air office - but for the right use cases it’s a great user experience. (Chris)
- There will be at least 6+ labs that release models more capable than o3; in practice, these models will still be too expensive to use for most applications, but will be used extensively to generate synthetic data sets and serve in reinforcement learning roles that will help train much smaller/faster/cheaper models. (Hugh)
- We see an emergence of “AI Relationship Coaches” that analyze communications in real-time and provide tailored suggestions for improving rapport and trust. (Salma)
- AI developer tools built around specific point solutions (e.g., RAG) will continue to fade while truly great platforms focused on reliability and evals will emerge. (Annie)
- There will be a <10B parameter open source model that performs better than Llama 3.1 405B; small fast models will be “good enough” that a large fraction of people relying on them will use local/on-device models for all their transparent/background tasks (like Copilot and Cursor autocomplete) and only make API requests for questions that take time to compose. (Hugh)
On AI within enterprises -
- Board-level conversations will shift from “What’s your AI strategy?” to “What results have you driven with AI?” Investments in workflow automation and well-designed product features will yield the best answers. (Chris)
- Every functional budget will include an AI line item. AI initiatives and funding will no longer be confined to an “Office of Innovation” or “AI Center of Excellence.” (Annie).
- We see a wild west of pricing for enterprise applications and services as the tangible value and ROI of these solutions increases while the cost to build and replicate them decreases. (Xerxes)
- We’ll see a resurgence of the UX function. As AI engineering skills become more commonplace, truly great AI-human design will be the differentiator between successful and unsuccessful workflow automation. (Annie)
- AI Product Ethicists will emerge as mandatory roles within PM teams, specifically focused on mitigation AI bias and ensuring responsible AI feature development across the product lifecycle. (Salma)
- Between cloud-native managed services (AWS Bedrock, Azure OpenAI) and increasingly sophisticated AI Governance policies, privacy and security concerns will stop being as big a bottleneck to AI adoption for many US companies. (Annie)
On AI’s cultural and societal impact -
- Europe will advance plans for state-structured AI, with a mandate to improve the lives of those being left behind and prevent economic disruption. (Joshua)
- There will be a risk of widening inequality between those with and without access to advanced AI tools. (Salma)
- While avatars for business use become passé, personified avatars of fan communities and other personal use cases start to show up. (Joshua)
- We'll see our first deepfake voice/video cloning political scandal, large-scale spearphishing, or police sting. (Joshua)
- We’ll see better accessibility to mental health support through AI-assisted services. (Salma)
2025 will continue the momentum of moving beyond the early AI hype toward real, measurable AI-driven results. While we anticipate significant advancements in AI models, the most interesting developments will likely come from how we put AI to work.
