Beyond the prompt: what real-time AI collaboration looks like

A new AI model demonstrated by the firm Thinking Machines Lab offers a glimpse of a more natural, collaborative human-computer interface. Unlike current chatbots that operate on a prompt-and-response basis, this new system can listen, watch, and respond in real time, even interrupting and being interrupted. This points towards a significant evolution in agentic AI, moving from discrete tasks to persistent, background assistance that could fundamentally change business workflows.

In a recent demonstration highlighted by the AI Daily Brief, a research group called Thinking Machines Lab showcased a new type of AI model designed for real-time collaboration. The key innovation is a move away from the rigid, turn-based interaction that defines current chatbots and AI assistants.

Instead of a user typing a prompt and waiting for a response, this model operates as a persistent, ambient partner. It can watch a user's screen, listen to their voice, and interject with relevant information or perform tasks in the background, all with very low latency. Crucially, the interaction is fluid—the user can interrupt the AI, and the AI can interrupt the user, much like a natural human conversation.

From turn-based to real-time

This represents a fundamental shift in the primary interface for interacting with AI. The current model forces humans to adapt to the machine's limitations, while the new model is designed to adapt to human patterns of work and communication.

This development suggests the industry is moving towards what podcast host NLW described as "what comes after chat"—a more integrated and natural form of human-AI partnership.

For business owners and operators, this transition from turn-based chatbots to real-time collaborative agents is not just an incremental improvement; it signifies a new paradigm for productivity and workflow design.

Redefining workflows and productivity

The primary impact will be the reduction of friction and context-switching. Today, using an AI assistant often means stopping your primary task, opening a separate chat window, crafting a detailed prompt, and then integrating the output back into your work. This process carries a significant cognitive load.

A real-time collaborative agent, however, could work alongside an employee within their existing tools:

• An analyst building a financial model in Excel could have an agent verbally fact-check figures or pull public data in real time, without ever leaving the spreadsheet.
• A customer service representative on a call could have an agent listening in, pulling up relevant knowledge base articles, and pre-filling ticket information in the CRM simultaneously.
• A designer working in Figma could ask an agent to generate ten variations of an icon based on the current canvas, with the results appearing instantly for review.

The goal is to move the AI from a tool you consciously go to into an assistant that is simply there, removing the bottleneck of the prompt-and-response cycle.

Implications for business operations

• Cost & Efficiency: By embedding AI directly into workflows, the time spent on manual data retrieval, context switching, and administrative overhead could be dramatically reduced. This leads to higher output per employee and potentially faster service delivery.

• Training & Onboarding: New employees could be paired with an AI agent that acts as a real-time coach, guiding them through complex software or internal processes and answering questions contextually as they work.

• Risk & Governance: The "always-on" nature of these agents presents new challenges. Businesses will need to establish clear rules for what data the AI can access, when it can operate, and how its actions are logged and audited. Data privacy and security become even more critical when an AI has persistent access to an employee's screen and microphone.

While this technology is still in its early stages, astute business leaders should begin preparing for its arrival now. The shift to collaborative agents will favour organisations that are agile and prepared.

1. Audit your current workflows for friction. Identify the specific processes where your team loses the most time to context-switching. Where do they have to stop what they're doing to look something up, ask a colleague, or use a separate tool? These high-friction points are the prime candidates for early adoption of collaborative AI.

2. Monitor the technology landscape. Keep an eye on developments from major labs like OpenAI, Google DeepMind, and Anthropic, as well as emerging players like Thinking Machines Lab. This is likely to be a major focus of research and product development over the next 12-24 months.

3. Experiment with 'precursor' tools. Get your team comfortable with AI-in-the-loop technologies that exist today. Tools like GitHub Copilot for developers or real-time meeting assistants like Otter.ai provide a taste of what more integrated AI feels like. Use these experiments to learn how your team adapts.

4. Strengthen your data governance. A collaborative agent is only as good as the data it can access. Start the work now to organise your internal knowledge, standardise your data formats, and establish robust security policies. A clean, well-structured data foundation is a prerequisite for effective and safe AI collaboration.

5. Plan for small-scale pilots. As these tools become commercially available, resist the urge for a big-bang rollout. Identify a single, measurable workflow and pilot the technology with a small, tech-forward team. Focus on clear metrics like task completion time, error rates, and employee satisfaction to build a business case for wider adoption.