The new AI race: Enterprise innovation in 2026

OpenAI Confirms Ads Are Coming to ChatGPT

The panel agrees that OpenAI’s decision to introduce advertisements into the ChatGPT experience is a pragmatic, if disappointing, step. The move is seen as an economic inevitability driven by the high cost of inference for its massive user base, estimated at 800 million weekly users. This follows a well-trodden path previously taken by search engines and social media platforms.

A key point of discussion is the potential impact on user trust, echoing the early days of Google when its founders argued that an ad-based model would create “perverse incentives” for a search engine. The panelists expressed concerns about how this would play out with a conversational AI, which is often perceived as a trusted assistant.

• Principles and Trust: OpenAI has stated principles, such as not allowing ad profiles to influence the generated content and keeping ads visually separate. The hope is that they maintain this “firewall” between the product’s core function and its economic model. The risk is a slide into an “influencer model,” where it becomes difficult to distinguish between genuine responses and paid promotions.
• User Experience: One panelist noted that a key reason for using ChatGPT over competitors was precisely to avoid influencing their advertising profile. The introduction of ads, even in the $8/month “Go” tier, marks a significant shift in the product’s value proposition.
• A Potential Upside: On a positive note, it was suggested that an ad-supported model could subsidize free access for students, educators, and others who cannot afford a subscription, thereby increasing accessibility.

Claude Code’s Breakout Moment and the Rise of Agentic AI

There is a consensus that Claude Code is experiencing a “ChatGPT moment,” where a sudden surge in public awareness and experimentation is revealing the true potential of agentic AI for coding. This is not due to a single new feature but rather the culmination of incremental model improvements and a more refined user experience that has crossed an “uncanny valley,” building a new level of trust among developers.

• From Assisted to Agentic: The key shift is from AI as an assistant (e.g., code completion) to a true agent that can be delegated complex software development tasks. Developers are moving from “babysitting every interaction” to a more hands-off approach, trusting the agent to work autonomously towards a solution.
• Simple is Powerful: It’s noted that Claude Code achieves this with a surprisingly small set of tools (e.g., file system access, git), demonstrating that the power lies in the core model’s reasoning and context management rather than an extensive ecosystem of plugins. The model’s ability to handle context effectively is its real key to success.
• Future Challenges: While the progress is significant, challenges remain. Key areas for improvement include more robust memory management to avoid context window limitations, better orchestration, and enhanced security. The evolution of these agents will also necessitate a change in the roles and workflows within the software development lifecycle (SDLC).
• Resetting the AI Race?: The panel considers the idea that Claude Code has “reset the AI race,” giving its creators an edge over competitors like OpenAI in the coding domain. The superior user experience and rapid iteration have captured significant developer attention.

Insights from the “Enterprise in 2030” Report

A new report from IBM’s Institute for Business Value (IBV) reveals a strategic shift in how executives view AI’s role in the enterprise. The findings suggest a move away from using AI purely for cost-cutting and efficiency towards leveraging it for top-line growth and innovation.

• From Productivity to Innovation: 64% of surveyed executives believe that by 2030, competitive advantage will stem from AI-driven innovation rather than just resource optimization. This marks a maturation of enterprise AI strategy, where the goal is not just to make the existing “steam train” run more efficiently but to fundamentally rethink the journey and create new value.
• Reinvesting Gains: 70% of executives plan to reinvest the value from AI-powered productivity gains into growth-oriented initiatives, signaling a commitment to using AI for strategic expansion.
• Core Components for Success: The report underscores that success requires more than just a powerful LLM. The critical factors are:
  • Grounding: Architecting systems to ground models in proprietary enterprise data and context.
  • Orchestration: Integrating and orchestrating AI agents within core business processes.
  • Skills Transformation: Recognizing that the shelf-life of technical skills is shortening and that future value will come from human creativity, ingenuity, and a mindset geared toward collaboration with AI. Knowledge worker roles are expected to be fundamentally transformed.
• The Boardroom vs. Reality Gap: While executives are bullish on this vision, a gap often exists between this high-level strategy and the on-the-ground reality. Employees in the business and IT departments understand the immense difficulty of this transformation and sometimes perceive a lack of consistent, top-down support to “walk the talk,” which is necessary to overcome the technical, operational, and governance hurdles.

Open Responses: A New Standard for Agent APIs

Hugging Face has introduced “Open Responses,” a new project aiming to create an open standard for agentic APIs. This initiative seeks to abstract more of the agentic process to the server side, representing the next step in the evolution of model interaction.

• The Evolution: The interface with models has evolved from raw numbers, to text (with the tokenizer on the server), to chat formats (with the chat template on the server). Currently, the client is still responsible for managing the agentic loop—interpreting a model’s request to use a tool, executing it, and returning the result. Open Responses proposes moving this entire agentic loop to the server. The client simply defines a goal, and the server handles all the intermediate reasoning and tool-use steps.
• The Good (Standardization): This move is an attempt to create a formal, open standard for agents, avoiding the fragmentation of the de-facto “OpenAI-compatible” chat APIs. It offers convenience for developers by simplifying the client-side logic.
• The “Dark Side” (Opacity and Protectionism): Moving more logic to the server allows model providers to hide more of their “secret sauce.” This is seen as a protectionist play, similar to OpenAI’s Assistants API, which obscures the underlying reasoning traces. Frontier model labs can use this to prevent their chain-of-thought processes from being observed and distilled into open-source models.
• The Trade-off for Innovation: This creates a significant trade-off. While it may be more convenient for users, it harms innovation in the open-source agent development community by reducing transparency. It slows the “cross-pollination of ideas” that comes from being able to see how frontier models reason and solve problems.
• A Contrarian View: One panelist strongly argued against this abstraction, stating that hiding the chain-of-thought is detrimental. Developers lose the ability to craft prompts that trigger the most efficient “circuits” within a model, effectively ceding control and insight to the model provider. This locks developers into a specific reasoning format, potentially disadvantaging other models and stifling the discovery of novel agentic techniques.

• Tags:
• Agentic AI
• ChatGPT
• Enterprise AI
• OpenAI
• Claude Code
• IBM
• Hugging Face
• API Standards