Source: raw/Anthropic_Just_Revealed_How_AI-First_Companies_Actually_Use_AI.md
Creator: Ben AI · URL: https://www.youtube.com/watch?v=-jl4-8nHdH4 · Duration: 14:51 · Platform: YouTube · Uploaded: 2026-09-16
Ben AI read through Anthropic’s library of customer case studies and pulled out eight adoption patterns, each applied to three readers: the business owner, the AI agency selling to that owner, and the employee trying to become the in-house “AI champion.” Every company example and number below is Ben’s paraphrase of a case study. None were checked against Anthropic’s published stories for this article, so confidence is low on the figures and higher on the patterns. Ben also sells a 30-day one-on-one coaching programme for managers, which strategy 1 promotes.
Key Takeaways
- Managers adopt first. Ben says IG Group and DoorDash found team-wide adoption was “significantly higher” where managers used Claude themselves than where teams relied on individual champions. His advice to managers is to block an hour a day for automating your own tasks; he expects AI to become your default way of working by week two or three. For agencies: train or win over the executive team before rolling out to staff.
- Don’t appoint AI champions; let them emerge. DoorDash, per Ben, gives everyone access and promotes whoever turns out to be a prolific user, because “you cannot really predict who your AI power users will be.” Tactics: a weekly AI stand-up where each person shows one use case, and the usage tracking in Claude’s Team plan.
- Put AI where people already work. One case study’s COO (company name garbled in the captions) had Claude run tasks overnight and post the results to a public Slack channel; the most common reply was a request for Cowork access, and Ben says the finance team has since automated more than 60 workflows. Ben recommends Claude Tag to bring Claude into Slack or Notion so people learn by watching power users.
- Connect the software. Ben says Workato saw employee adoption rise by more than 700% after connecting Claude to Salesforce, Gong and internal tools through MCP. To roll out connectors without sharing credentials, he recommends Composio: a meta-connector the owner connects once, giving each employee scoped access through a single connector, with per-department permissions on its business plan.
- Build a company second brain. Sierra, Zapier and HubSpot each keep a shared context repository (customer insights, decisions, data analysis, brand and voice guides). Ben’s is a folder of text files any teammate can select in their AI tool; he says one person can set it up in a couple of hours. For agencies, setting up and maintaining one is a service you can sell.
- Automate the work nobody gets to. Ben argues the largest wins came from work that wasn’t being done at all. Examples he cites: a two-year backlog project (company name garbled) that Claude finished in 20 hours, and Duvo’s retail monitoring workflow, previously run “maybe quarterly at best”, which he says saved €1.45M in its first week. His own team’s examples: following up a year of unworked CRM leads, and a customer-intelligence report built from support calls.
- Anyone who does a task three times builds a skill. The process owner knows the edge cases, so they should build the skill, not IT. Ben’s examples: Blank Metal (30 people), whose non-technical CRO built a lead-processing skill in an hour that scores each lead, creates the HubSpot deal and posts to Slack; and Brainlabs (about 1,000 people), which reached around 400 skills in four weeks. Record a Skill lowers the bar. For agencies: build alongside the process owner, “because if they can’t run and change it after you leave, it won’t really stick.”
- Skills are team assets. Ben quotes HubSpot’s rule: “if I’ve built a workflow that saved me 30 minutes a day, my whole team probably also needs it.” On the Team plan, Customize → select skill → Share sends it to named people. Updates sync to every recipient, recipients can’t edit, and the creator stays the owner, so make the owner responsible for keeping it working.
Try It
- Start with the leadership team. If you are the agency, the first deliverable is one hour of hands-on work with the owner or practice manager, not a staff rollout.
- Run a weekly 15-minute AI stand-up where each person shows one use case. Note who keeps showing up with new ones.
- Post one useful AI output to a shared Slack channel each day before asking anyone to adopt a tool.
- Connect the three systems people copy and paste from most (CRM, email, docs) before training anyone on prompting.
- Create a shared context folder (ICP, brand voice, offers, strategy) and point every teammate’s Claude at it.
- List three jobs that never get done (old leads, review responses, lost-deal follow-up) and automate one.
- Adopt a three-times rule: any task done three times becomes a skill, built by its owner and shared from the Team plan.
Open Questions
- Every figure is Ben’s paraphrase, not checked against the case studies: Workato 700%+, Duvo €1.45M in week one, Brainlabs ~400 skills in four weeks, the 20-hour backlog project, 60+ finance workflows. Check the original Anthropic customer stories before citing any of them.
- Two company names are caption-garbled (“leak’s” COO; “Whis” backlog project) and could not be identified.
- Skill-sharing behaviour (recipients can’t edit, updates auto-sync) is Ben’s description of a Team-plan feature; not confirmed against Anthropic docs here.
- Composio’s business-plan permissions are as Ben describes them; no wiki article on Composio exists to check against.
Related
- AI Transformation Case Studies from The Artificial Intelligence Show — executive first-hand rollout accounts
- 2026 — How AI Is Restructuring Organizations and Knowledge Work — the organizational picture these tactics sit inside
- Claude Tag — Claude inside Slack, strategy 3
- Record a Skill — non-technical skill building, strategy 7
- Distributing Team Skills as a Plugin Marketplace — another way to share skills across a team, strategy 8
- Ben’s Five-Skill AIOS Setup — Ben’s second-brain build, strategy 5
- The 5 AI Automations Businesses Actually Pay For — what agencies sell once adoption is in place