Last year, we were consumed by the potential of generative AI.
Flash forward to today, where AI continues to be on the tip of our tongues for every discussion, but it’s agentic AI that’s now stealing the show. Over the past year, we’ve observed incumbent leaders like Salesforce and ServiceNow launch AI agents and agentic workflow platforms, in addition to start-ups like Clay and Decagon emerging with best-of-breed AI apps and agents.
At G2, we are not only focused on expanding our AI taxonomy to chart where the market is heading, but we also offer our own Specialized AI Agent, Monty, for Sales.
Given the timeliness of this trend, I was ecstatic to host Tomasz Tunguz, General Partner at Theory Ventures, at our G2 Reach event to discuss the future of agents—from the players to watch to the most impactful use cases driving major productivity gains.
Here are some of my key takeaways from our discussion.
The startup advantage
While early AI advantages went to large incumbents thanks to their vast data resources and PhD talent pools, Tomasz made a case for why startups will ultimately win in the AI agent space. His belief hinges on the premise that startups' agility, propensity for risk, and rapid adaption to changing user experiences and workflows provide a distinct edge over larger organizations often laden with legacy systems.
Sales development representatives (SDRs) are a prime example, where traditional software was built around managing a single inbox with perhaps 50 outbound messages. In the AI-enabled future, SDRs will effectively manage 50 inboxes simultaneously. This workflow shift creates an opening for startups to innovate while incumbent platforms remain constrained by their legacy architectures, Tomasz noted.
The use cases fueling early success
When I asked Tomasz about the impact of AI agents to date, he shared the impressive results being observed already across multiple domains. In customer support, he shared deflection rates have jumped from 15-25% with previous-generation tools to 60-70% with current AI solutions.
Even more, these improvements aren't limited to simple queries like password resets, as AI agents are successfully handling complex technical support cases. In one example, Tomasz noted insurance companies repurposed 70-90% of their data processing teams after implementing AI solutions that can accurately parse complex documents like W2s and utility bills.
He also highlighted marketing and sales use cases with personalized messages and accurate forecasting, noting the high potential in these areas. He shared: “I think you'll see account-based marketing delivered at infinite scale. That's incredibly effective...for the content generation, but also for the account research, specificity in delivering the messages.”
Overcoming the trust paradox
One important theme of our conversation was around building trust in AI agents. Tomasz drew an interesting parallel to self-driving cars. While human drivers cause tens of thousands of fatalities annually, we hold autonomous vehicles to a much higher standard of near-perfect safety, he noted.
While human accuracy in many tasks hovers between 65-75%, AI systems likely need to achieve close to the 85-95% accuracy mark before widespread trust is established. An opportunity to help us get there involves using multiple AI models to cross-check each other's work, similar to how companies use multiple human reviewers for critical tasks.
What's next for AI agents (and those who embrace them effectively)
While we've seen significant progress in AI's capabilities, Tomasz believes we're entering a new phase where the focus will shift from raw model performance to innovative combinations of specialized models working together.
As inference costs continue to drop and new chip architectures emerge (including in our phones and laptops), we'll see increasingly sophisticated AI agent workflows that can handle complex tasks with higher reliability than ever before.
For enterprise software leaders, the message is clear: the next few years will bring new opportunities to reimagine workflows and create value through AI agents. The winners will be those who can most effectively harness these technologies while building the trust necessary for widespread adoption.
If you missed my conversation with Tomasz, you can catch it on-demand, along with all Reach sessions.
