BAG Ventures, founded by former Google executive Bonita Stewart and former CapitalG partner Jackson Georges Jr., has closed an $11.3 million fund to invest in early-stage AI companies. Its thesis is straightforward: enterprise AI buyers are becoming less interested in pilots and general-purpose chatbots, and more interested in software with clear economics, workflow integration and a short route to revenue.
The firm has already invested in 10 companies, including software company SXD, AI travel agent BizTrip and agentic reasoning platform Nomadic. It plans to deploy the remaining capital over roughly the next two years, writing checks of $100,000 to $500,000.
The investment signal: AI must earn its place in the workflow
BAG is targeting AI infrastructure, compute, physical and edge AI, security, governance and vertical SaaS. But the more consequential filter is commercial rather than technical.
Georges argues that enterprise procurement is moving beyond an “experimental sandbox” phase. Buyers are scrutinizing unit economics and favoring deterministic products that can execute defined work inside existing processes—such as code review automation or legal-document processing—rather than offering an open-ended conversational interface.
For operators, this is a useful framing. An AI product’s appeal increasingly depends on whether a business can identify:
- a discrete job the system can complete;
- a measurable cost, speed or quality improvement;
- the systems and approvals required to deploy it; and
- a credible owner for the resulting outcome.
That favors startups built around a specific operational bottleneck over broadly positioned AI assistants. It also raises the bar for founders: BAG says it looks for teams with a minimum viable product, at least one partner and a clear path to monetization within 24 hours.
A go-to-market network is part of the product
The fund’s claimed advantage is access to potential enterprise customers. Stewart and Georges previously ran the BAG Collective angel syndicate, which has more than 450 members. BAG Ventures says it has more than 150 limited partners, including Google and operators from companies such as Nvidia, Amazon and Snowflake.
The firm intends to pair capital with warm customer introductions and hands-on go-to-market advice. That matters because selling AI into established organizations remains difficult even when a product performs well in a demo. Security review, data access, workflow ownership, integration work and budget approval can each stall adoption.
For founders, investor networks may be most valuable when they help validate a buying process—not simply generate introductory meetings. A useful customer connection should help test whether the product has an identifiable budget, deployment path and accountable business owner.
Defensibility shifts from the model to the operating context
BAG is also explicitly avoiding startups that are merely thin layers on top of frontier-model APIs. As foundation-model providers release more capabilities and applications themselves, a technically competent interface may not remain differentiated.
Instead, the firm is seeking companies embedded in enterprise workflows that can capture proprietary, non-scrapable data and retain what Georges calls the “intent layer.” In practical terms, that means understanding how work moves through an organization: the policies, exceptions, approvals, permissions and historical decisions that turn a model response into an action.
Regulated industries are a particular focus. Their requirements around privacy, internal data flows, acceptable-use guardrails and continuous automated red-teaming can create both implementation friction and a moat for vendors that meet them.
What to watch next
BAG’s thesis points to several tests for the next enterprise AI cycle. Watch whether buyers increasingly contract around completed tasks or business outcomes rather than per-user software seats; whether agentic systems can meet reliability and governance requirements in production; and whether startups can establish defensible workflow and data advantages before model providers commoditize their core features.
Another emerging category is identity and access management for AI agents. As non-human systems gain permission to retrieve data and trigger actions, enterprises will need controls tailored to agent identities, authority and auditability. BAG sees this as an opening for new zero-trust and orchestration infrastructure.
The central takeaway is less about one new fund than a changing purchase criterion: enterprise AI companies will need to prove they can safely deliver a business result, not just demonstrate that a model can generate an answer.
