AI in enterprise and startups

From: aidotengineer

The 2025 AI Engineer World’s Fair highlights the profound impact of AI across both burgeoning startups and established enterprises. Speakers emphasized that the current wave of AI innovation is a genuine revolution, distinguishing itself from past technological hype through widespread, practical adoption [00:17:48].

The AI Revolution: Beyond Hype

Unlike previous fads such as blockchain, NFTs, or the metaverse, the current AI surge is characterized by tangible products that people are actively using [00:17:59]. ChatGPT, for instance, achieved 100 million users faster than any consumer product in tech history, with millions engaging daily to accomplish tasks [00:18:02]. This widespread utility, even for questionable uses like cheating on essays or generating fake legal citations, underscores its real-world impact [00:18:13].

AI in Enterprises

Major technology companies are deeply embedding AI into their core offerings and strategies:

• Microsoft: GitHub Copilot boasts millions of subscribers, and Copilot is now integrated into Microsoft 365, serving 84 million consumers [00:18:35]. Azure AI alone generates $13 billion in annual revenue from enterprise adoption [00:18:46]. Microsoft’s goal is to empower every person to use AI to shape the world [00:36:42].
• Amazon Web Services (AWS): AWS plans to spend $87 billion on AI infrastructure this year, indicating a massive commitment to the space [00:19:10].

Enterprise adoption is driven by the maturation of AI models, which can now generate hypotheses, understand unstructured data, and perform at a PhD level in specific domains [00:37:37]. These models are also becoming more efficient, moving from data centers to local devices, offering greater control and lower latency [00:37:54].

Key shifts in enterprise AI strategies include:

• From Pair Programming to Peer Programming: Tools like GitHub Copilot are evolving from sidekicks to actual teammates in development workflows [00:38:24].
• From Software Factory to Agent Factory: The focus is shifting from shipping binaries to shipping agents that can retrain, redeploy, and adapt post-launch [00:38:33].
• Local AI: Models are increasingly running on devices and at the edge, crucial for industries like healthcare due to compliance and privacy requirements [00:57:10]. An example is a hospital system using local AI to summarize longitudinal patient data [00:57:35].

Microsoft’s Foundry platform supports this shift, enabling a “signals loop” where models are fine-tuned with real-world interactions for dramatic quality improvements [00:45:52]. Dragon, a healthcare co-pilot, achieved an 83% character acceptance rate after fine-tuning with 650,000 interactions [00:46:11]. Foundry also features “agentic RAG” (Retrieval-Augmented Generation), which improves accuracy by 40% on complex queries [00:47:37].

AI in Startups

The scale of AI startups is “impressive, but has no visible means of support,” metaphorically compared to the largest pillarless ballroom west of Las Vegas [00:15:57]. This highlights both the rapid growth and the inherent risks.

• Rapid Growth: Companies are achieving $10-100 million in Annual Recurring Revenue (ARR) faster than in any previous technology revolution [01:04:31].
• Successful Examples:
  • Llama Index: Touted as a leading framework for building AI applications [00:16:10].
  • Cursor: An AI-powered developer productivity platform, reaching millions to $100 million ARR in 12 months with minimal sales effort [01:12:19].
  • Cognition: An AI agent that is a top committer in many companies [01:12:31].
  • Windinsurf: An agentic IDE acquired by OpenAI for $3 billion [01:12:39].
  • Hey Gen and 11 Labs, Midjourney: Multimodal AI companies rapidly exceeding $50 million ARR [01:08:21].
• The “AI Leapfrog Effect”: Surprisingly, conservative and low-tech industries are adopting AI fastest [01:17:11]. Examples from Conviction Capital’s portfolio include:
  • Sierra: Resolves 70% of customer service queries for clients like SiriusXM and ADT [01:17:20].
  • Harvey: Essential for competitiveness in the legal industry, well over $70 million ARR [01:17:31].
  • Open Evidence: Helps doctors stay updated with medical research, used weekly by a third of US doctors [01:17:41].
• Execution as a Moat: The key to success for AI startups is execution, not necessarily inventing new models or product surfaces [01:22:01]. Building “thick” products—comprehensive workflow wrappers—that collect and package context, use the right models, and present thoughtful outputs, is crucial [01:15:10]. This requires deep domain knowledge and understanding of user workflows [01:15:47].

Key Trends and Challenges

• Model Cost Reduction: The cost of powerful models has plummeted dramatically, making AI broadly accessible [01:10:51]. GPT-4, for example, dropped from $30to$2 per million tokens in 18 months [01:10:51].
• Multimodality: The ability to process voice, video, and images alongside text is expanding the potential impact of AI across vast sectors of the economy [01:08:11].
• Agentic Evolution: The industry is moving beyond co-pilots to more autonomous agents that can hold goals in memory, try hypotheses, and backtrack [01:06:55].
• The Model Context Protocol (MCP): An open-source, standardized protocol for models to interact with the outside world, enabling model agency and wider adoption of tool calling [02:34:21]. MCP aims to simplify integrations and empower builders to create agents that interact with various services seamlessly [02:35:15].
• Challenges in Implementation:
  • “MCP is just another API wrapper” syndrome: A common pitfall where developers simply expose existing APIs as tools, leading to poor AI performance [03:05:41]. It’s crucial to design for how agents and models consume context [03:26:18].
  • Security Concerns: Giving AI systems access to private data and malicious instructions poses significant risks, including prompt injection [01:42:30]. Strong security measures and careful control over tool access are paramount [02:57:18].
  • Client Support and Standardization: While MCP is gaining traction, full client support for its advanced features (like sampling and elicitation) is still evolving [03:06:08].

In conclusion, the AI landscape is undergoing a fundamental transformation, offering immense opportunities for both startups and enterprises. The key lies in understanding the evolving capabilities of AI, focusing on real-world problems, and executing with a user-centric, workflow-driven approach, rather than merely adopting the latest technology [01:14:01].