1. AI tools are gaining deeper, more persistent access to how we work.

The launch of Anthropic's Cowork in early 2026, in our view, highlighted a meaningful shift in direction: AI tools are moving beyond discrete applications that employees open when they need them toward products with broader, more persistent access to files, calendars, messages and context.⁽¹⁾ We believe this shift could have implications for how knowledge work gets done.

Cowork's rapid evolution, from research preview in January 2026 to a full enterprise product with connectors spanning finance, legal, HR and operations by February, represents one example of this shift, but Anthropic is not alone.  We’ve seen Microsoft's Copilot agents, Salesforce's Agentforce and Google's AI-embedded tools all moving in the same direction. Across the industry, we see AI tools being designed to maintain richer, more continuous access to the systems and information people work with every day, a shift that is moving quickly across functions.

For leaders, we believe the near-term priority is less about adopting Cowork or other similar products and more about working to ensure your organization has the practices and foundations in place to work effectively with AI as it becomes more deeply embedded in daily work.

2. Prompting has evolved.

For the past couple of years, getting reliable output from AI typically meant crafting long, detailed prompts, specifying exactly what to do, how to do it and what to avoid. While that approach can work, it is fragile and can be difficult to scale. Prompts are often hard to reuse, challenging to share across teams and require maintenance as models and business needs evolve.

A potentially more durable approach is emerging: reusable, structured instruction files that describe a goal and the expected output, rather than scripting every step. We believe the practical benefits are worth understanding. These instructions can produce more consistent and repeatable results, help facilitate collaboration across tools and teams, and even be generated and refined by AI models themselves, making it easier to standardize how work gets done with AI across an organization.

Anthropic formalized one version of this approach in December 2025, with the release of Skills as an open standard. In our view, this release signaled a move toward more portable, shareable AI instructions that can extend beyond a single platform and become an increasingly important part of enterprise AI infrastructure.

The tools are still maturing, and enterprise-wide management and distribution remain a work in progress. While it is still early, we believe Skills and other reusable, structured instruction approaches offer meaningful promise and are worth exploring as organizations look for more repeatable ways to standardize, structure and scale their AI workflows.

3. Context is key to unlocking AI potential.

As AI models become more widely available and capable, we believe proprietary business context has emerged as a key to unlocking AI’s potential within an organization. Teams that want to expand AI use cases to help address more complex tasks, in our view, will likely need more than what can be typed into a prompt or found on the open internet. They need context: relevant information about the company, the individual managing the project, internal practices and policies, and enterprise data from internal documents, CRM systems, data warehouses and other enterprise sources that serve as a source of truth for the business.

We see context management as a central challenge for organizations today, both assembling the right context and providing that context to models in a secure and repeatable way. AI models can work with both the structured data found in databases and unstructured data found in documents, emails and multimedia; however, this can also create data quality challenges that are broader than many organizations have historically managed. Approaches like Retrieval Augmented Generation (RAG), Model Context Protocol (MCP) and native connectors provide programmatic access to context, and we believe they are steps in the right direction. However, in our view, connecting AI models to enterprise data in a way that is accurate, secure, practical and repeatable remains difficult.

We continue to believe that investments in data infrastructure, integration and governance are foundational to realizing the potential of AI. For business leaders, the gap between generic AI output and meaningful AI output is often a context problem in disguise, and as AI becomes more deeply embedded in how work gets done, the cost of that gap grows. We believe companies that have built strong data foundations will be better positioned to take advantage of context-aware AI, and that head start will grow more valuable as the tooling, standards and market solutions in this space continue to mature.

4. Broad AI adoption presents a leadership challenge.

A common AI adoption challenge we hear from leadership teams is human rather than technical. Despite real investment in AI tools, we’ve seen that usage often clusters among a small group of enthusiasts while the rest of the organization remains skeptical or simply hasn't figured out how to incorporate them into their daily work. Broad AI adoption, therefore, presents a leadership challenge; in our experience, the organizations pulling ahead have made adoption both an organizational and a technical priority.

Software engineering is further along the adoption curve than most other functions, and we believe it offers an early instructive use case for leaders working toward broader AI adoption throughout their organization.⁽²⁾ Even for companies that invested in broad deployment across the engineering organization, it took time for models and tools to mature to the point where the average engineer could reliably integrate them into daily work and drive measurable productivity gains.⁽²⁾⁽³⁾ While this challenge is common with the adoption of many new technologies, we believe AI adoption is further exacerbated by the "jagged frontier" of the models – the uneven, unpredictable and shifting capabilities many models exhibit.

To help address this, and support the adoption urgency felt across many functions, leaders can encourage bottom-up experimentation to help surface use cases and also emphasize the importance of embracing AI from the top down. In our experience, those executives who use AI tools themselves, talk openly about what they're learning and share where they're facing challenges, see better traction. Peer networks can be combined with more formal training; in some cases, people may move faster when they hear from peers and colleagues about what's actually working. Organizations can also turn to consultancies and to the AI labs themselves, both of which are launching solutions focused on helping enterprises climb the adoption curve. Ultimately, we believe the organizations that succeed will be those that recognize the full scope of what broad adoption requires, recognizing it as both an organizational change and a technology deployment.

‍