For the past several months, Microsoft Copilot has been our primary AI assistant across the firm. That made sense as a starting point: it lives inside the Microsoft 365 tools our team already uses every day, it came with predictable flat-fee pricing, and it required little change management to ensure the team was consistently augmenting its work with AI for everyday tasks. In Copilot we have a low-friction way to expand our internal AI fluency and it delivered.

That calculus has since changed. I thought sharing what changed would be both consistent with our commitment to transparency and may be useful to others facing a similar choice.

Generative AI, tools that produce content from a prompt, was the right frame for where the technology stood a year or two ago. It is no longer the frame that matters. The more consequential shift is toward agentic AI: systems that take action inside real workflows and existing software, not just answer questions about them. For a firm that builds and implements systems for mission-driven organizations, this is the layer where AI actually starts to change how work gets done.

Microsoft’s restructured pricing gave us a second reason to re-evaluate the market. Over the last few months, much of the new agentic capability in both Microsoft 365 Copilot and GitHub Copilot has moved behind feature-based and usage-based pricing rather than the flat subscription we started with. Agentic features cost more to run, and that cost is now being passed through.

That cost model creates a problem for anyone investing heavily in agentic workflows. We would need either a significantly larger license commitment or an open-ended usage-based agreement—with no clear ceiling on cost. Worse, it penalizes exactly the behavior we have built into our service delivery model, where routine tasks are automated and skilled operators lean into this technology to solve complex problems.

We also noted that Microsoft Copilot runs on a mix of OpenAI and Anthropic models under the hood. That prompted an obvious question: if we already rely on those models indirectly, why not work with them directly?

After evaluating the field, we selected Claude as our preferred AI solution and Anthropic as our preferred AI partner going forward. A few reasons stood out.

• It is incredibly powerful general-purpose AI. Anthropic is one of only two U.S.-based companies developing genuinely frontier-class AI models, and Claude consistently ranks at or near the top of the benchmarks relevant to our work.

• The pricing model fits how we work. Like OpenAI, Anthropic prices primarily on usage (API tokens) rather than layering feature-gated licenses on top of a subscription. Our plan provides all staff with access to the full range of Claude’s functionality, with token usage as the only meaningful constraint. We believe that model is easier to forecast, control, and explain to our clients than a landscape of stacked feature tiers. It also more directly reflects what these companies are actually charging for: the cost of building and running frontier models, rather than the cost of feature development layered on top of them.

• AI safety + values matter to us. Anthropic also has the most explicit public commitment to AI safety among the major model providers, and has taken public positions that are more likely to align with the values of the organizations we serve. As a Certified B Corp, we weigh that alignment as much as raw capability.

We know many of our clients are still cautious about AI. That caution is about social and environmental impact and responsible use, not just capability or cost. We share those concerns. Part of why we have chosen Anthropic is that its stated safety commitments and research priorities give us more confidence in recommending Claude to the organizations we serve.

We will continue to be direct with clients about the tradeoffs involved in any AI deployment, including compute and energy considerations, as we help them think through their own adoption. We remain steadfast in our commitment to carbon neutrality as a company and will factor AI use into our plans.

One other benefit: Anthropic offers meaningful nonprofit pricing. Those discounts should help ensure Claude is a responsible recommendation to those organizations typically most resourced constrained in our portfolio.

This shift is additive, not a replacement of the partnerships that already define our work.

Salesforce remains at the center of our practice, as it has for the past decade-plus. That does not change. We will continue to build and manage CRM environments on the platform, and we are enthusiastic about what Agentforce will offer the sector. Claude is complementary to that work, not competing with it. In practice, using Claude to deliver Salesforce services lets us build better, more elegant solutions on the platform—often faster and more efficiently.

Microsoft also remains an important part of our technology stack. Most of the organizations we support run Microsoft 365. Microsoft Power BI, Microsoft Azure, and Microsoft Fabric remain important tools for advanced analytics, custom app development, and data management across many client architectures. While we are gradually phasing out Copilot internally, it will remain among the tools we support for clients as part of Managed Services.

Across both platforms, we think of Claude and Anthropic as the layer that strengthens our overall technology strategy, for us and for the clients we support.

We are building internal expertise on Claude and Anthropic’s platform for that reason. Our clients are already asking about Claude, and we intend to be a firm that not only uses Claude expertly as part of our own service delivery, but also helps organizations adopt it responsibly themselves. We expect to stand as a recognized Anthropic partner in this space later this year.

We will have more to share on what this looks like in practice for our client engagements in the coming months.