Salesforce acquires Convergence: How it competes with OpenAI, Anthropic, and Google’s Duet AI

Salesforce’s acquisition of London-based AI startup Convergence represents a significant turning point in enterprise automation. Announced on May 15, 2025, this purchase strengthens Salesforce’s Agentforce platform, positioning it as a formidable competitor to OpenAI’s Operator, Anthropic’sClaude, and Google’s Duet AI.
Convergence was founded in April 2024 by two seasoned machine learning engineers:Marvin Purtorab(CEO) andAndy Toulis(CTO). Both previously worked at Shopify, where they focused on recommender systems and AI assistants, before contributing to Cohere, a leading enterprise AI platform. Since launch, the company has assembled an impressive team from top AI research institutions, includingGoogle DeepMind,Meta,OpenAI, and PolyAI.
In September 2024, the startupsecureda $12 million seed round led byBalderton Capital, with Shopify Ventures andSalesforce Venturesparticipating as strategic investors. Vladyslav Yatsenko, co-founder and CTO ofRevolut, joined as a shareholder. This early funding demonstrated strong investor confidence in the team’s vision and capabilities.
The acquisition announcement came on May 15, 2025, just one year after Convergence’s founding. Balderton Capital achieved a remarkable 300% IRR on its investment, as noted in a since-editedLinkedIn postby Balderton partner James Wise. Industry analysts value the deal between $150–$200 million, reflecting the high demand for premier AI agent talent.
Convergence’s Large Meta Learning Model (LMLM) architecture is central to the acquisition. This technology enables AI agents to adapt in real time to web interfaces, handling pop-ups, errors, and UI changes while maintaining long-term memory across tasks. Through continual learning, the system enhances agent capabilities and coordinates multi-agent workflows for collaborative problem-solving.
These features give flagship product of Convergence, Proxy, significant advantages over competing platforms. Beta users report completing tasks 40% faster than with Anthropic’s Claude and requiring 60% fewer manual interventions than OpenAI’s Operator. This superior performance stems from LMLM’s unique training approach, emphasising learning ability over task execution, mimicking human cognitive adaptability.
Beyond technology, this acquisition secures crucial AI engineering talent. Convergence’s 150-person team, including 12 former Google DeepMind researchers and 8 Meta AI veterans, will form the foundation of Salesforce’s new London AI Lab. This move aligns with Salesforce’s $4 billion UK technologyinvestmentand helps counter the talent exodus to U.S. tech companies.
Jayesh Govindarajan, Salesforce’s EVP of AI/ML Engineering, emphasised London’s strategic value: “With Europe’s densest cluster of AI PhDs and proximity to Cambridge’s ML research hubs, this lab will spearhead our next-generation agent development.” The move also leverages the UK government’s £1 billion AI infrastructure fund announced in March 2025.
With the acquisition, Salesforce has overhauled Agentforce’s pricing structure. The previous model charged $2 per conversation with a 50-action limit. The new Flex Credits system, launching May 2025, offers 100,000 credits for $500 ($0.10 per action), 20 credits per action (such as updating CRM records or managing support tickets), and unlimited internal agent usage at $550 per user monthly.
This pricing makes Agentforce 60% cheaper than OpenAI’s Operator ($1,200 per 100k actions) and 45% less than Anthropic’s enterprise API. Early adopter OpenTablereportsa 35% cost reduction in customer service automation since switching to Flex Credits.
Salesforce plans to integrate Convergence’s technology across three strategic areas. First, they’ll enhance the Agentforce ecosystem. In Q3 2025, industry-specific AI agents will be introduced for healthcare (automating prior authorisation) and finance (assisting with loan underwriting). By Q4 2025, these agents will integrate with Slack workflows, enabling simple commands like “/book_meeting” for seamless team automation.
Second, Salesforce will develop a hybrid reasoning framework by combining its Einstein large language models (LLMs) with Convergence’s large meta-learning models (LMLMs). This fusion will create AI agents with enhanced contextual decision-making, such as CRM-based sales lead prioritisation, and self-correction capabilities for marketing workflow errors.
Third, Salesforce is preparing for upcoming regulatory requirements, particularly enforcing the EU AI Act’s 2026. They’re implementing audit trails for agent decisions and bias detection tools, including benchmarks from models like Claude 3.7, to ensure transparency and compliance.
This strategic move has prompted responses from competitors: OpenAI is accelerating its EU Operator platform launch (previously set for Q3 2025), Anthropic is partnering with AWS to cut Claude API costs by 30%, and Google is expanding Duet AI beyond Workspace to SAP and ServiceNow. Yet Salesforce maintains unique advantages: 1.5 million Einstein AI-trained developers, native CRM integration reducing latency, and Flex Credit subsidies through their Foundations program.
Technically, Salesforce Agentforce now leads in real-time web adaptation and collaborative workflows, while offering competitive pricing and robust compliance features. Its hybrid LMLM-LLM architecture and deep Slack/CRM integration distinguish it from the more siloed approaches of OpenAI, Anthropic, and Google.
As Balderton’s James Wise noted, this acquisition transcends a successful investor exit — it marks Europe’s declaration of independence in the global AI agent race. With the deal’s completion expected by July 2026, Salesforce has initiated a new enterprise automation leadership contest phase.