Google Cloud Summit: Enterprises Are Ditching Chatbots for Autonomous AI Agents
Maureen Costello of Google Cloud says 25 percent of organizations already use agentic AI, with adoption expected to hit 75 percent within two years.
The annual Google Cloud Summit in London has made one thing clear: enterprise AI strategy is moving on from chatbots. The new priority is autonomous agents, systems that can set goals, reason through multi-step tasks, and call external APIs to complete work independently. Maureen Costello, Google Cloud vice-president for UK, Ireland, and sub-Saharan Africa, noted that the term was barely on the agenda a year ago; it is now the central focus of enterprise transformation.
Adoption Is Moving Fast
Survey data presented at the summit shows that one-quarter of organizations currently use agentic AI at a moderate level. That figure is projected to triple to 75 percent within the next 24 months. Leaders from OVO Energy and insurer Hiscox described how agents are already cutting customer call volumes and reducing underwriting times from hours to minutes by pulling information from across internal silos simultaneously.
Four Pillars for Getting It Right
Google executives and industry specialists outlined four priorities for deploying agentic AI successfully. The first is scope: build tools that solve specific problems, such as client onboarding, rather than constructing sandboxes with no clear outcome. Felix Reilly of the UK government's Incubator for AI argued that a tight focus on outcomes builds internal trust, because staff can see direct benefits to their daily work.
The second priority is cost control. AI services are priced per token, and unpredictable spend can stall scaling plans. Alex Rutter, EMEA managing director of AI at Google Cloud, recommended using a mix of models. Routine tasks can be handled by traditional machine learning algorithms with near-zero compute costs, while complex reasoning is reserved for more capable systems like Gemini.
The third pillar is talent. No-code tools let almost anyone design an agent, but connecting those agents to older, complex data sets requires deep technical expertise. Organizations without that capability in-house are increasingly relying on forward-deployed engineers from model providers, including Google and Anthropic.
The fourth is stability in production. Experts at the summit recommended building at least 20 to 30 benchmark tests to monitor accuracy continuously. They also advised reviewing thought logs and transcripts regularly to catch circular reasoning, where an agent loops without resolution and burns through expensive tokens in the process.
What This Means for Africa
Costello's remit covers sub-Saharan Africa, and her framing of the agentic shift as something happening now carries weight for the region. In Nigeria, Kenya, and South Africa, businesses frequently contend with legacy infrastructure and fragmented data systems. Agents offer a path to automating complex workflows without requiring a full overhaul of existing technology, a meaningful advantage where capital for large-scale transformation is limited.
The tokenomics conversation is especially relevant for African founders working with tight budgets and currency volatility. The summit's advice to route routine tasks through lighter, cheaper models and reserve high-cost compute for critical reasoning maps directly onto the constraints many local tech firms already face. As Google deepens its focus on the sub-Saharan market, applying agentic tools in African finance, retail, and logistics could lower operational costs meaningfully while improving service reach for users who have historically been underserved.
The window for experimentation is closing; organizations that are not building real agentic tools today are already behind the companies that are.
Source: Wired
Written by
StacyAI-assisted news curation. Every story is reviewed by our editors before publication.
