A snapshot from the end of 2024
At the end of 2024, TechCrunch reporter Rebecca Szkutak asked 20 venture capitalists who back enterprise startups what they expected in 2025. At the time, enterprise AI adoption had been slower than many hoped. Budgets were tight and AI often stayed in the experimental category.
The predictions are worth revisiting now because they captured the moment when attention shifted from chatbots to agents. This post groups the main themes and draws practical lessons for teams putting agents to work.
From chatbots to agents
Aaron Jacobson of NEA predicted that AI budgets would be allocated away from chatbots and toward agents, with enterprises moving beyond "GPT wrappers" to "deploy digital workers that can reason and take action to make a real business impact."
Ed Sim of Boldstart Ventures thought about second order effects. If people soon have dozens or hundreds of agents working for them, he asked, who provides access control, who manages those agents, and is there a platform to manage and secure many different agents at once?
Kathleen Estreich of Pear VC noted that adoption had lagged because use cases were not tightly scoped and tools to reduce errors and check outputs were not yet strong enough. She expected vertical agents built for specific end to end workflows to be a large opportunity.
Data comes first
Several investors stressed data. SC Moatti of Mighty Capital said AI adoption hinges on better data, especially as companies move from experiments to large scale use. Jonathan Lehr of Work-Bench pointed to data sovereignty, meaning control over where data lives and how it is processed. Scott Beechuk of Norwest argued that the deepest moats would come from large proprietary datasets.
Time to first value
Mike Hayes of Insight Partners highlighted a metric he expected to matter more: time to first value. He described it as a proxy for how easy a product is to implement. Faster time to value, he said, should be an advantage.
Marell Evans of Exceptional Capital was watching how long enterprises trial tools before adopting them and how AI pricing compares with traditional software, including usage based and outcome based models.
Where investors saw opportunity
The areas investors named included code agents for modernizing older applications, observability and incident response, IT service management, sales engagement, security operations, task specific models paired with agents, and services businesses where AI could automate behind the scenes work.
Hayes also described a shift from thinking of technology as either driving revenue or cutting costs toward technology that reduces friction across the business, including workflows reimagined with agentic automation and security tools that fix problems instead of only raising alerts.
How investors judged a moat
Cathy Gao of Sapphire Ventures described a framework built on design, data, domain expertise, distribution and dynamism. Raviraj Jain of Lightspeed suggested a simple test: as models get better, does this company get threatened or strengthened?
Budgets and adoption
Most investors expected budgets to hold or rise, with a focus on measurable returns. Jason Mendel of Battery Ventures cited a Battery Ventures poll of 100 CXOs in which 74 percent expected to increase their technology spend in 2025. Hayes expected boards to push for AI use cases in production. Mark Rostick of Intel Capital noted that it was easier to adopt AI through application vendors than to build a platform, and saw pent up demand for platform solutions.
What holds up today
Looking back, several themes have aged well. The move from chatbots to agents that take action is now the center of enterprise AI conversations. The questions Sim raised about managing and securing many agents have become practical concerns. And the emphasis on tightly scoped use cases, good data and fast time to value matches what teams learn when they move agents into real work.
What did not age as well
Not every prediction played out evenly, and the source itself shows a range of views. Some investors expected enterprise adoption to jump quickly, while others expected budgets to stay flat for part of the year. The lesson is that adoption tends to follow proof. Organizations move faster when a use case is narrow, measurable and low risk.
Lessons for teams building with agents
Scope tightly. Pick specific workflows with clear outcomes rather than broad assistants.
Get your data in order. Agents are only as good as the information they can reach.
Plan for many agents. Decide early how you will control access, monitor activity and manage changes.
Measure time to value. Choose approaches that show results quickly, then expand.
Keep people in the loop. Use review steps where accuracy matters until results are reliable.
Where Innflow fits
Innflow is an all in one workspace for AI agents and workflows. It speaks directly to the question Ed Sim raised about managing many agents in one place. In Innflow, teams build agents and workflows side by side, connect them to their tools, and manage them from a single workspace rather than across separate products.
That supports several of the lessons above. You can start with a narrow workflow to get to value quickly, add agents where judgment helps, control what each one can reach, and add approval steps where a person should decide. As the number of agents grows, you can follow their progress from one workspace.
From prediction to practice
Many of these predictions described a future of digital workers that reason and act. For many teams, that future is now a planning question. If you want one workspace to start building and managing agents and workflows together, visit innflow.ai.
Source: Rebecca Szkutak, TechCrunch, "From AI agents to enterprise budgets, 20 VCs share their predictions on enterprise tech in 2025". Read the original.












