Good expert call on Anthropic with a Head at Cognizant Cognizant ran GitHub Copilot, concluded it was not working, evaluated alternatives, and moved to Claude Code. That was announced late last year and is now rolled out broadly rather than sitting behind special approval. His view is that snapshot benchmarks are not the basis for a decision like this, because model leadership rotates. What drove it was that Anthropic is climbing the enterprise value chain faster than OpenAI, which he characterizes as having been model-first and then consumer-first. For a system integrator, the B2B orientation is the product. There is also a specific technical reason tied to their core delivery work. Cognizant does heavy legacy modernization, pulling apart COBOL mainframe estates. Claude Code's larger context window lets them work on bigger chunks of a migration at once, which is the binding constraint in that work. "we are just half a step ahead of the customers and maybe not even that" Developers are running 30-35% more effective. That is spreading into testing, and then into business analyst work, requirements capture, prototyping and UI drafting, with architecture behind that. The commercial problem is that an integrator on time and materials does not automatically capture its own efficiency gains. He places this inside a 10-15 year erosion of pricing power: clients have become far more professional buyers, and the platform companies are taking a growing share of total IT spend. Outside of engineering he describes it as very early. Most people are drafting emails and getting help with presentations. The step change he describes is a project brain: an entire RFP, with notes, templates and background, loaded into a single Claude project that a whole team works against. There are also hard limits on what can be uploaded, and customer contracts cannot go in. He is specific about what agentic automation needs before it clears the bar for regulated back-office work: governance, happy path analysis, human-in-the-loop, auditability for BPO processes, confidence scores, and an ontology over the underlying documents so that a policy document is parsed the same way every time. His view is that Anthropic does not have that functionality yet. For live BPO work they use a native agentic automation platform instead, Otera, which is Austrian and fully EU-resident including data centers and personnel. He expects Anthropic to get there on the trajectory of the last six months. The gating factor is not the software. "the end state will make the current consumption look like nothing... it's a business transformation, and it needs to be driven like one" He explicitly guards against recency bias. Everything has gone Anthropic's way for five or six months and he thinks the market is over-extrapolating that. His view is that Anthropic is pulling ahead but it is early, that OpenAI will have a real enterprise play, and that the space comfortably supports two primary providers. He is more concerned about the second tier, naming Meta, falling behind. source: Tegus