Two Google alumni have closed an $11.3 million fund to back early-stage AI startups, betting that the era of enterprise AI experimentation is ending and that customers will increasingly pay only for products that prove their worth.<\/p>\n
BAG Ventures, founded by Bonita Stewart, a former Google vice president, and Jackson Georges Jr., a former CapitalG partner, closed the fund after about two years of investing from it as it came together. The firm has already backed 10 companies, including the software company SXD, the AI travel agent BizTrip, and the agentic reasoning platform Nomadic.<\/p>\n
It invests in startups in areas like AI infrastructure, compute, physical and edge AI, security, governance, and vertical SaaS. Check sizes range from $100,000 to $500,000, and the team hopes to invest the rest of the fund over the next two years.<\/p>\n
Stewart spent 17 years at Google, including nearly a decade as a vice president. During that time, she also served on the board of Gradient Ventures, Google\u2019s early-stage AI fund. She is a limited partner in the Female Founders Fund and the Operator Collective. With Georges, she also co-led the angel syndicate BAG Collective, which has more than 450 members.<\/p>\n
Georges worked at GE Healthcare and at Google, where he met Stewart. He later became a partner at CapitalG, Alphabet\u2019s growth fund. He and Stewart were in the first cohort of the Black Venture Institute at Berkeley.<\/p>\n
The pair say their edge is access. They launched BAG Ventures to \u201caddress the emerging AI divide between founders and operators,\u201d Georges said.<\/p>\n
\u201cFounders needed inside access to the organizations they wanted to sell into, and we knew so many high-level operators who wanted to support early founders but didn\u2019t know how,\u201d he said. As a result, \u201cwe don\u2019t just give founders capital; we give them direct warm introductions to potential customers and hands-on [go-to-market] advice,\u201d he continued, adding that the firm, whose limited partners include Google, as well as operators from Nvidia, Amazon and Snowflake, has more than 150 limited partners altogether at a wide range of companies.<\/p>\n
\u201cLots of firms have an operator network,\u201d Georges said. \u201cWe want to be one that\u2019s genuinely useful.\u201d<\/p>\n
Georges\u2019s investing thesis rests on a shift he sees in how enterprises buy AI. He said the \u201cexperimental sandbox\u201d phase is ending. \u201cEnterprises are dialing in heavily on the unit economics right now,\u201d he noted. \u201cThey aren\u2019t just paying for open-ended chatbots anymore; they are paying for deterministic solutions. The real value is coming from solutions that integrate deeply into legacy workflows and actually execute the work.\u201d Examples include automating code reviews and parsing legal documents.<\/p>\n
Georges is preparing for a world where enterprises no longer buy per-user seats for SaaS tools. \u201cWe\u2019ll be buying completed jobs and outcomes driven by multi-agent workflows,\u201d he said.<\/p>\n
Toward that end, BAG Ventures wants core technical teams that have worked together before, have a minimum viable product and at least one partner, and have \u201ca very clear path to monetization within 24 hours.\u201d<\/p>\n
He also wants to back founders building products that go deep into enterprise workflows and capture proprietary data that can\u2019t be scraped. With frontier AI labs launching more products themselves, not even a technically sound product from a startup is enough to succeed in the long run otherwise, he observed, and \u201cif a startup is just a thin wrapper around a frontier model API, they\u2019re going to get wiped out.\u201d It\u2019s why they look for teams building products that go deep into enterprise workflows and capture proprietary data that can\u2019t be scraped. \u201cWe want companies that own the intent layer and have the customer lock-in to survive the next big model release.\u201d<\/p>\n
The firm is also looking at startups selling into highly regulated industries, where data privacy needs may require specialization. \u201cThat means securing internal data flows, building acceptable-use guardrails, and deploying continuous automated red-teaming,\u201d said Georges. \u201cWe\u2019re already seeing this approach work well with our portfolio company Defendremate.\u201d<\/p>\n
Georges also said enterprises will need Identity and Access Management tools for non-human workers, like AI agents. \u201cStartups that can build the next level of \u2018zero trust\u2019 architecture and orchestration rails specifically for agentic systems are going to fill a massive and very lucrative gap,\u201d he said.<\/p>\n