AnthropicA verified review on Soft Atlas

Anthropic review

Arnaud G., Mom3nt5/5

Quick review
Arnaud G.

Expert admin·Mom3nt·France

5
July 2026

I run a 3-person B2B SaaS startup in Paris. I am the CEO, not an engineer, and Claude is the reason my sales operation runs like a team of six. We use Claude inside our product, running different Claude models depending on the task, the heavier reasoning on one, the faster generation work on another. Choosing per use case rather than defaulting to the biggest model is what made the economics work. The part I did not expect is that Claude also replaced my ops stack. I built a family of Skills in Cowork that runs two full workflows. The meeting prep routine reads my calendar, decides which meetings actually deserve a briefing, pulls the account history from my CRM, my inbox and my call recordings, checks the prospect's recent LinkedIn activity, and produces a pre-filled discovery brief delivered before the call. Thirty minutes of manual prep per meeting, down to zero. The outbound routine chains targeting criteria, pain mapping, voice guidelines and account research into a runnable prospecting workflow across my sales tools, then reads back what actually converted and returns ranked recommendations. The architectural decision that made Skills work: separating doctrine from execution. Who we sell to, what we sell against and how we write live in read-only skills that never change during a run. The orchestrators consume them. That is what stopped the system from drifting after a few weeks, and I would not have found it without Claude pushing back on my first design. What I would improve. MCP connector reliability is uneven, some servers drop mid-run and the failure is silent rather than explicit. Skills lack observability: when a chain of four skills produces a bad output, I cannot see which one degraded. Long unattended runs still need more supervision than I want to give them. Usage limits are the real constraint on running these workflows daily at the scale I want. None of that changes the rating. This is the only tool in my stack I could not replace.

More reviews of Anthropic