By Arnab Bose, Chief Product Officer, Asana.getty​Most companies have already crossed the AI-adoption threshold, but the challenge that remains is execution.​AI can generate content, code and plans, but in most enterprises, these tools still operate outside the living context of how work actually gets done. They are missing key information about who owns what, what depends on what, what matters now and why certain decisions were made. That is why there is a gap in which individual AI usage is rising far faster than companywide productivity gains. According to Asana's research, 75% of knowledge workers use AI, yet only 5% of companies see meaningful productivity improvements.​​The companies that get more from AI will not be the ones with the most prompts. They will be the ones that give AI the context to operate inside complex workflows, with shared memory, clear accountability and governance.​Chat is a useful interface, but a poor operating model.Different types of work have different requirements, and people naturally reach for different modes accordingly: chat for quick guidance, drafting and simple requests, or structured tasks and workflows for work that needs to persist, be tracked or handed off.​​Chat took off by being fast, intuitive and easy to get started. It offered an on-ramp for AI, one that helps people generate outputs. However, it is not the right long-term interface for managing complex work.​Chat starts to fail at enterprise scale when teams need to manage large volumes of AI-generated work together. As agents produce more assets, tasks and recommendations, the bottleneck shifts from output to coordination. ​Work does not scale in disconnected threads. Teams need visibility into what is happening, clarity on who owns what and reliable handoffs between people and systems. That is why the long-term opportunity in enterprise AI is not another chat surface. It is bringing AI into the flow of execution.​Enterprise AI needs shared context, not isolated prompts.​Much of the conversation around agentic AI centers on increasing autonomy, yet independent action without sufficient context introduces significant risk. According to Asana data: "73% of ITDMs say AI initiatives fail or stall at least sometimes due to incomplete organizational context. Moreover, 43% of workers spend more than 30 minutes per day fixing or reworking AI outputs because key context was missing.​" The research was conducted by Censuswide among 1,002 IT decision-makers aged 18-plus in the U.K. and U.S., excluding sole traders, and 3,002 knowledge/white-collar workers aged 18-plus in the U.K. and U.S.The next wave of productivity gains will come from giving AI access to a shared system of work: an enterprise context layer that turns disconnected prompts into coordinated work by providing an accurate context layer and tells AI who is involved, what matters, what depends on what and how work actually gets done.​In many organizations, work is scattered across chats, tools and one-off automations with no shared memory. Instead, organizations can ground workflows in a shared knowledge base that captures previous decisions and established ways of working. That foundation gives every new interaction a stronger starting point while allowing the AI to improve through accumulated experience. Over time, it creates lasting organizational knowledge rather than forcing the system to begin from scratch with each exchange.​​The future of work is multiplayer by default.The most important design question in enterprise AI is no longer what one person can ask a model to do. It is how humans and agents work together inside the same system.​The shift underway is from AI as personal assistance to AI as part of execution. That means moving beyond co-pilots and productivity tools toward agents working alongside people on the same shared plan to move work forward with clear identity, permissions and human oversight.​We already know agents can handle repeatable, template-driven work. What they cannot replace is human judgment: taste, prioritization and the ability to make decisions in context. The best enterprise systems will protect and amplify those distinctly human strengths through effective human-agent collaboration.​The future of AI at work is visible, governed and built for execution.The challenge in enterprise AI is no longer generating output. It is coordinating and executing work at scale. As agents move from assistance to execution, companies need a system where work is visible, accountable and governed. Not scattered across disconnected threads.​That is why enterprise AI needs more than chat. Work has to live inside the flow of execution, where tasks, decisions, dependencies and handoffs can be tracked, reviewed and coordinated across teams. Moving forward, the most valuable work systems must allow users to move seamlessly between different modes without losing context, ownership or history.​We know that structured environments with explicit ownership, visibility and handoffs generally outperform unstructured communication for complex, interdependent work. That’s why agents should function like true collaborators inside that system, with shared memory, clear guardrails and checkpoints for human judgment. This is what makes AI durable inside a business: not a side tool, but part of a governed, multiplayer system for getting work done.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?