AI agent startups raise funding for enterprise workflows

Recent reports describe a wave of funding for startups that put AI agents into enterprise workflows, and Google Cloud has introduced agent products for finance and legal teams. The reported rounds include $100 million for groundcover, $150 million for HappyRobot, $63 million for Skan AI and about $300,000 for Zenalyst. The tools automate calls, emails, documents, research and workflow steps inside existing systems, but most adoption numbers come from the companies themselves.

Observability for agents

VentureBeat reported that groundcover announced a $100 million Series C led by One Peak, taking total funding to $160 million. Groundcover gives developers visibility into how AI agents behave. It runs in the customer's own cloud, uses eBPF to capture telemetry without per-gigabyte charges, and includes an Agent Mode that lets engineers ask questions about incidents in natural language. The company says it has more than 250 paying customers and tripled annual recurring revenue, though those figures are company-reported.

Workflow execution and context

Tech.eu reported that HappyRobot raised a $150 million Series C at a $1.2 billion post-money valuation, with Prysm Capital leading and Eurazeo co-leading. Participants include a16z, Base10, Y Combinator and Koch Disruptive Technologies. HappyRobot's agents carry out work across voice, email, documents and the web inside supply chain and enterprise operations. It names DHL, Kuehne + Nagel and Uber as customers and says one customer automates 28,000 hours per month, while customer care agents resolve more than 70 percent of requests autonomously. Those metrics come from the company.

citybiz reported on August 12 that Skan AI raised a $63 million round co-led by Cathay Innovation and Dell Technologies Capital. Skan watches how employees work across enterprise applications, builds a context graph, and then uses that context to run AI agents. Skan says revenue grew more than 300 percent, net dollar retention reached 150 percent, and seven of the ten largest U.S. banks are customers. At one bank, it measured 11.2 million context switches and says agents cut cost per transaction by 32 percent, lifted throughput by 41 percent and produced $18 million in annualized savings. The company reports these results.

Ascendants reported that Zenalyst raised Rs 3 crore, about $300,000, in a pre-seed round from angel and corporate investors. Its ZenForce platform includes ZenBank for treasury, ZenProcure for procurement and ZenLegal for contract intelligence. Zenalyst says its agents connect to more than 150 enterprise systems, reduce manual effort by up to 90 percent on target workflows, and some deployments pay back in under 12 months. Its sales pipeline is about $1 million, according to the report.

Google enters regulated industries

ITWeb reported that Google Cloud launched Gemini Enterprise for Financial Services and Gemini Enterprise for Legal. The finance product has a financial research agent, more than 50 skills and 13 data connectors to sources such as FactSet, LSEG and SEC Edgar. It handles KYC research, credit risk assessment and bond issuance, and outputs include citations and confidence scores. The legal product handles contract review, due diligence, regulatory monitoring and privacy requests, and connects to tools such as Thomson Reuters, iManage and RelativityOne.

Where the skepticism comes from

Even with this activity, enterprise adoption is not yet proven. Gartner projects worldwide AI spending of $2.59 trillion in 2026, up 47 percent from 2025, but also predicts that more than four in ten agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear value and weak controls. Carter Page, EVP of R&D at Astronomer, wrote in CIO that organizations can see their total AI bill but often cannot trace which workflows generated it. The FinOps Foundation says controlling token costs is the top concern for practitioners.

Companies can reduce costs by redesigning workflows rather than just choosing cheaper models. OpenAI, for example, discounts repeated input tokens by up to 75 percent when applications use prompt caching. The startups described above sell different layers: groundcover sells visibility, HappyRobot and Zenalyst sell workflow execution, Skan sells a context layer, and Google sells regulated industry agents. Which deployments produce real savings remains open, and the next stage will depend less on funding announcements than on whether companies can trace the cost of each automated task and show the result was worth it.