Windsurf, BUZZ HPC, Gnosis, ElevenLabs, Prosus & Bloomberg: The Agentic Economy AI’s Unseen Hand

Panelists and Core Businesses

• Tim Cowley (BuzzHPC): CFO of a major NVIDIA-certified GPU cloud provider operating in Canada and Sweden since 2017; 100% green energy data centers; offers an "AI refinery" for tuning open-source models.
• Jero (Prozus): Lead AI for Prozus, a consumer internet group serving 2 billion users globally across 100 countries; listed on the Amsterdam Stock Exchange.
• Alex (Windsurf, formerly Codium): Field CTO building the first AI IDE; serves 54% of the Fortune 500 in regulated industries (finance, insurance); acquired by OpenAI for $3 billion.
• Unidentified Speaker (Gnosis): Integrates AI agents into blockchain for prediction markets, trading, fraud detection, and user protection.
• Matty (11 Labs): Co-founder of a voice AI company focused on multimodal interaction; created a marketplace with over 5,000 voices where creators have earned $5 million.

Definition and State of AI Agents

• An AI agent is defined as autonomous software that perceives its environment, makes decisions, and takes actions to achieve preset goals with feedback loops.
• Successful agents require continuous or semi-continuous operation to learn from feedback and adapt behavior in similar future situations.
• Key capability requirement: Agents must operate across modalities, including voice, visual understanding, text, and coding.
• Adoption drivers include improved model reasoning capabilities, reducing error compounding in multi-step problems to make production deployment viable.

Operational Metrics and Enterprise Case Studies

• Prozus: Reduced hallucination rates from 10% in 2022 to below 1% by introducing agents to create trustworthy, multi-step conversations.
• Prozus Workforce: Released internal tools enabling employees to build agents; 500 "agentic colleagues" were created within three months to automate specific workflows.
• DHL Implementation: Agents enabled autonomous decision-making for warehouse capacity and shipping optimization.
• DHL Results: Delivery times improved by 20% and warehouse costs reduced by 15% over the last 12 months attributed to agentic solutions.
• OpenAI Acquisition: Agreed to acquire Windsurf for $3 billion, citing the high traction of AI agents in software engineering due to clear compile/run feedback loops.

Adoption Trends and Use Cases

• Coding: Primary use case due to the binary nature of code (works/fails), extensive open-source training data (GitHub, Stack Overflow), and high engineer experimentalism.
• Rapid Prototyping: Tools like "Lovable" allow sales and non-engineers to prototype full applications quickly by combining voice, LLMs, and code.
• Prediction Markets: Agents summarize research to predict future outcomes; monetary incentives drive users to correct errors, creating an infinite improvement loop.
• Bottom-Up Innovation: Organizations achieve better results when employees build their own agents rather than top-down mandates, facilitating collective learning.

Technical Challenges and Constraints

• Data Disambiguation: Systems struggle to map ambiguous natural language queries to precise database facts; requires 99% accuracy for utility versus 80% which is considered useless.
• Knowledge Uncovering: Major effort required to formalize organizational vocabulary and structure unstructured databases (e.g., inconsistent column naming) for agent consumption.
• Hallucination vs. "I Don't Know": LLMs are trained to output answers rather than admit ignorance, necessitating robust fallback mechanisms and testing against "edge case" queries.
• Human-in-the-Loop (HITL): Critical for managing latency and determining when to pause for human review; tools must facilitate efficient acceptance/rejection of changes.
• Integration Complexity: Standardizing authentication and workflow integration with existing enterprise systems (CRMs, legacy tools) is identified as the next major evolution.

Security, Ethics, and Regulation

• Voice Impersonation: Requires preemptive user disclosure and future audio watermarking to authenticate agents and prevent misuse.
• MCP Authentication: Current Multi-Context Protocols (MCP) face scalability bottlenecks in authentication; resolving this is critical for agent autonomy.
• Regulatory Compliance: EU AI Act and California Consumer Discretionary Bot Act mandate disclosure when interacting with AI agents.
• Strategic Transparency: Disclosing AI usage builds customer loyalty; non-disclosure risks legal liability and brand damage.
• Functional Guardrails: Hard-coded checks must prevent agents from executing high-risk actions (e.g., unrestricted money transfers) without specific authorization.
• Exception to Disclosure: Agents may be used undetected in anti-fraud scenarios (e.g., wasting time with scam callers) to protect users.

Future Infrastructure and Internet Evolution

• Task-Based Internet: Shift from navigation-based browsing to task-oriented interactions where the internet acts as a backend API for agents.
• Advertising Model: Transition from Cost Per Click (CPC) to outcome-based and success-based pricing models.
• Agent Discovery: Companies must optimize for "agent discovery" rather than human search engine optimization (SEO).
• Operating System Shift: Agents will increasingly make decisions across OS, browser, app, and URL layers, reducing the need for manual step-by-step user input.
• Agent-Optimized Content: Websites may need to be specifically structured (e.g., APIs, structured text) for agents to process efficiently.

Implementation Strategy for Organizations

• Pilot Approach: Start with low-risk, high-value "low-hanging fruit" use cases like information retrieval or internal training within 1–3 months.
• Change Management: Critical success factor; requires educating staff on AI concepts (e.g., prompt engineering) to bridge the gap between Fortune 500 executives and frontline workers.
• Resource Allocation: Invest in IT infrastructure and staff training; consider hiring consultants for complex integrations.
• Evaluation Ratio: Expect a 75% evaluation and 25% automation split in the first year of deployment.
• Sector-Specific Risks: Finance and high-security sectors require reaching 99% accuracy before full deployment; lower-risk sectors can deploy earlier.

Workforce Impact and Economic Outlook

• Job Transformation: AI agents will replace some jobs but primarily augment human productivity; the primary risk is people with agents replacing people without them.
• Workforce Augmentation: Prozus estimates AI assistants generated 1.7 million hours of work in one year, equivalent to 1,000 full-time employees without added cost.
• Software Engineering ROI: AI allows firms to hire more engineers to tackle tech debt and feature creation by increasing the return on investment for developer salaries.
• Salesforce Benchmark: 40% of work at Salesforce is now performed by AI, primarily in software engineering and customer support.
• New Revenue Streams: 11 Labs' voice marketplace has paid $5 million to 5,000+ creators, enabling new income sources for voice talent via interactive agent experiences.
• Long-term Outlook: AI capabilities will drive revenue growth and new product offerings that will likely outweigh job displacement impacts in the 2–5 year horizon.