Claude+44 Positive

Operator Insights

What are business and product leaders learning from rolling Claude out across their teams?

10 accounts: 3 run it beside another tool, 1 evaluated it, 3 adopted it and 3 reported by others

Run it beside another tool3

James ReggioCTOBrex
Organization-wide. Fintech company. Described January 17, 2026.
Latent Space, published Jan 17, 2026Listen to the episode (opens Latent Space in a new tab)
Observation
Brex offers Claude beside ChatGPT and Gemini and lets each employee choose.
Takeaway
Buy a few seats of each and let use decide before an enterprise contract.
What happened
Brex buys a small number of seats of several assistants and lets employees request a Claude, ChatGPT or Gemini licence through ConductorOne in Slack, so it can switch as the products change.
“And so, for instance, like we allow employees to... basically go to, in Slack and use Conductor One to get a ChatGPT, a Cloud or a Gemini license.”
James Reggio, at 21:15 in the recording
- Podcast episode (Latent Space)
Deployment details
Organization-wide. Fintech company. Described January 17, 2026.
- Podcast episode (Latent Space)
Team
Employees who request a licence
Workload
Everyday assistant work
Product version
Claude, ChatGPT and Gemini licences

TinyJul 3, 2025Lenny's PodcastTiny's co-founder rates Claude best for writing and keeps ChatGPT and Gemini for other work.
Andrew WilkinsonCo-founderTiny
Individual. A holding-company founder's own stack, described July 3, 2025.
Lenny's Podcast, published Jul 3, 2025Listen to the episode (opens Lenny's Podcast in a new tab)
Observation
Tiny's co-founder rates Claude best for writing and keeps ChatGPT and Gemini for other work.
Takeaway
Give your writers a Claude seat even where another assistant is the default.
What happened
Wilkinson runs Claude beside ChatGPT and Gemini: Claude for writing, which he uses extensively and rates the best, Gemini for very large inputs and ChatGPT for everything else.
“Claude, I use for writing very extensively. I find it's the best for that.”
Andrew Wilkinson, at 51:25
Podcast episode (Lenny's Podcast)
Personal use
Deployment details
Individual. A holding-company founder's own stack, described July 3, 2025.
Podcast episode (Lenny's Podcast)
Personal use
Team
One founder
Workload
Writing
Product version
Claude, ChatGPT and Gemini
MetaJan 18, 2026Lenny's PodcastA Meta product manager uses Projects in both assistants and opened a Claude project when Meta approached him about a role.
Zevi ArnovitzProduct managerMeta
Individual. Personal projects, described January 18, 2026.
Lenny's Podcast, published Jan 18, 2026Listen to the episode (opens Lenny's Podcast in a new tab)
Observation
A Meta product manager uses Projects in both assistants and opened a Claude project when Meta approached him about a role.
Takeaway
Keep separate work in separate Claude projects so context stays where you put it.
What happened
Arnovitz uses ChatGPT Projects and Claude Projects as shared folders of chats with instructions and knowledge, and says ChatGPT's memory mixed unrelated contexts. When Meta approached him about a role, he opened a new project in Claude straight away.
“GPT projects and Claude projects, which are basically a shared folder of chats which share both custom instructions and shared knowledge base.”
Zevi Arnovitz, at 8:04 in the recording
Podcast episode (Lenny's Podcast)
Personal use
Deployment details
Individual. Personal projects, described January 18, 2026.
Podcast episode (Lenny's Podcast)
Personal use
Team
One user
Workload
Personal projects and job-application preparation
Product version
Claude Projects and ChatGPT Projects

Evaluated it1

Ethan MollickProfessorWharton School
Individual. A Wharton professor's own Gmail and teaching work, July 2026.
One Useful Thing, published Jul 23, 2026Read the write-up (opens One Useful Thing in a new tab)
Observation
Ethan Mollick tested Claude Cowork against ChatGPT Work on the same seminar task, and Cowork asked before sending.
Takeaway
Set what each agent may send on its own before you compare them, and run the same task on both.
What happened
Mollick asked both agents to connect to his Gmail, prepare an MBA seminar with slides and answer outstanding messages. Both built teaching material. Cowork drafted a reply for him to approve, as he had told it to; ChatGPT, which he had let send email, sent one.
“To use the computers provided by the AI companies, the mode you want is called ChatGPT Work in ChatGPT, and Cowork in Claude (the naming will not get less confusing, I am sorry to say).”
Ethan Mollick
- Public practitioner write-up
- Single hands-on test
Deployment details
Individual. A Wharton professor's own Gmail and teaching work, July 2026.
- Public practitioner write-up
- Single hands-on test
Team
One user
Workload
Seminar preparation from Gmail, with replies
Product version
Claude Cowork and ChatGPT Work

Adopted it3

HUB InternationalCo-issuer of a joint release with AnthropicHUB International
Organization-wide. Insurance brokerage. Deployment began in late Q4 2025, announced February 25, 2026.
PR Newswire, published Feb 25, 2026Read the report (opens PR Newswire in a new tab)
Observation
HUB International began rolling Claude out to its 20,000-plus employees and reports 2.5 hours saved a week in early use.
Takeaway
Ask any vendor-backed figure for its method; HUB's 2.5 hours a week comes from early use cases in a joint release.
What happened
HUB International began deploying Claude across its workforce of more than 20,000 employees in late Q4 2025. In a press release issued jointly with Anthropic on February 25, 2026, it reports early results of 2.5 hours saved per employee per week and over 90% satisfaction across early use cases, with no method stated.
“Early deployment results show 2.5 hours saved per employee per week and 90%+ user satisfaction from employees across early use cases…”
Hub International Limited
- Vendor customer story (joint press release with Anthropic)
Deployment details
Organization-wide. Insurance brokerage. Deployment began in late Q4 2025, announced February 25, 2026.
- Vendor customer story (joint press release with Anthropic)
Team
More than 20,000 employees
Duration
From late Q4 2025
Measurement
Hours saved per employee per week (2.5) and satisfaction (over 90%), early use cases

Yale UniversitySep 15, 2026Yale University (AI at Yale)Yale gave faculty and staff Claude Enterprise, with a separate HIPAA environment for patient data.
AI at YaleUniversity IT and AI servicesYale University
Organization-wide. Yale faculty and staff. Launched September 15, 2026.
Yale University (AI at Yale), published Sep 15, 2026Read the write-up (opens Yale University (AI at Yale) in a new tab)
Observation
Yale gave faculty and staff Claude Enterprise, with a separate HIPAA environment for patient data.
Takeaway
Ask Anthropic for the usage limits your licences carry before you budget seats, since Yale could not publish them.
What happened
On September 15, 2026 Yale gave faculty and staff Claude under a university-managed enterprise agreement, at US$20 or US$50 a month per licence, with Claude Chat, Code and Cowork. A separate HIPAA-compliant environment with Chat and Code serves people working with ePHI or data that needs zero retention. Yale notes that Anthropic does not publish usage limits for the licences, which run on a rolling five-hour window and a weekly cap.
“Faculty and staff now have access to Claude under a university-managed enterprise agreement.”
AI at Yale
Public organization write-up (university IT announcement)
Prices stated by Yale
Deployment details
Organization-wide. Yale faculty and staff. Launched September 15, 2026.
Public organization write-up (university IT announcement)
Prices stated by Yale
Team
Faculty and staff who buy a licence
Duration
From September 15, 2026
Workload
Chat, code and Cowork; a HIPAA environment for ePHI
Product version
Claude Enterprise, with a HIPAA-ready environment
PostHogApr 24, 2026PostHog blogPostHog runs a weekly Cowork check across about fifty teams' plans, and finds Cowork still bad at most writing.
Charles CookAuthor of the PostHog postPostHog
Professional team. Product analytics company of many small teams. Account published April 24, 2026.
PostHog blog, published Apr 24, 2026Read the write-up (opens PostHog blog in a new tab)
Observation
PostHog runs a weekly Cowork check across about fifty teams' plans, and finds Cowork still bad at most writing.
Takeaway
Hand Cowork scheduled checks over structured sources, and keep writing and browser tasks with people for now.
What happened
Every Monday at 08:08 London time, a scheduled Cowork task pulls the quarterly objectives of PostHog's roughly fifty small teams from the website repository on GitHub and looks for conflicts between them. Cook connects Cowork to PostHog, GitHub, Slack, Granola, Asana and Google Calendar, and says it is still bad at most writing, browser navigation and Google Drive.
“Cowork is still bad at some obvious things (most writing, navigating browsers, Google Drive).”
Charles Cook
Public engineering write-up
One operator's setup
Deployment details
Professional team. Product analytics company of many small teams. Account published April 24, 2026.
Public engineering write-up
One operator's setup
Team
One operator across about fifty small teams
Duration
Weekly, ongoing at April 2026
Workload
Reading team objectives in GitHub for conflicts; promise tracking
Product version
Claude Cowork with MCP connectors

Reported by others3

JamfApr 29, 2026Anthropic, Deploying Claude across your organizationJamf's HR team turned its performance review spreadsheet into a Cowork skill without waiting on engineering.
Matt BenyoDirector of AI InitiativesJamf
HR team. Cowork skill. Performance reviews.
Anthropic, Deploying Claude across your organization, published Apr 29, 2026Read the write-up (opens Anthropic, Deploying Claude across your organization in a new tab)
Observation
Jamf's HR team turned its performance review spreadsheet into a Cowork skill without waiting on engineering.
Takeaway
Turn a process into a skill only after the team has run it by hand enough times to know what good output looks like.
What happened
Jamf's performance reviews lived in a spreadsheet with seven competency facets and branching logic by level and role. The usual route to a usable tool was a quarter of engineering time and a custom internal app. HR had run the process by hand often enough to know what a good review looked like, so it encoded the process as a Cowork skill in 45 minutes. The skill asks the manager questions, applies the rubric for the role, and produces the review.
Vendor customer story (Anthropic deployment guide)
Build time reported
Deployment details
HR team. Cowork skill. Performance reviews.
Vendor customer story (Anthropic deployment guide)
Build time reported
Team
Jamf HR
Duration
45 minutes to build the skill
Workload
Guided performance reviews from a seven-facet rubric
Product version
Cowork skill
Measurement
Build time against an estimated quarter of engineering time
ZapierApr 29, 2026Anthropic, Deploying Claude across your organizationZapier's product marketing team drafts homepage messaging in Cowork from a skill that holds its positioning rules.
Joe StychHead of Product MarketingZapier
Product marketing team. Cowork with a custom skill and internal tools over MCP.
Anthropic, Deploying Claude across your organization, published Apr 29, 2026Read the write-up (opens Anthropic, Deploying Claude across your organization in a new tab)
Observation
Zapier's product marketing team drafts homepage messaging in Cowork from a skill that holds its positioning rules.
Takeaway
Write the team's standards into one skill so every homepage draft starts from them, and review the draft before design work begins.
What happened
Zapier's product marketing team gives Cowork the live homepage, a custom skill with its voice, positioning, and page-structure rules, and internal tools through MCP, including Slack threads and Glean search. Claude reads the page, identifies its modules, and returns an HTML draft of the page for the new positioning. Stych shares a draft with the team after about 15 minutes, before design or engineering is involved, and keeps working while it runs.
Vendor customer story (Anthropic deployment guide)
Time to first draft reported
Deployment details
Product marketing team. Cowork with a custom skill and internal tools over MCP.
Vendor customer story (Anthropic deployment guide)
Time to first draft reported
Team
Zapier product marketing
Workload
Homepage messaging drafts as HTML pages
Product version
Cowork with a custom skill and MCP connections
Measurement
About 15 minutes to a shareable draft
Thomson ReutersApr 29, 2026Anthropic, Deploying Claude across your organizationThomson Reuters skeptics came around on Cowork after a couple of hours of real work.
Joel HronCTOThomson Reuters
Claude Enterprise. Cowork trial with staff outside software development.
Anthropic, Deploying Claude across your organization, published Apr 29, 2026Read the write-up (opens Anthropic, Deploying Claude across your organization in a new tab)
Observation
Thomson Reuters skeptics came around on Cowork after a couple of hours of real work.
Takeaway
Put skeptics to work on real workflows with their own documents, and keep a person validating the output.
What happened
Thomson Reuters runs Claude Enterprise and piloted Cowork in teams that do not usually build software. Anthropic's guide reports that non-developers moved into automation and light prototyping with the documents and data they already use. A couple of hours on real workflows changed skeptics' minds. Hron describes the human role as validation, refinement, and decisions.
Vendor customer story (Anthropic deployment guide)
Pilot described, no figures reported
Deployment details
Claude Enterprise. Cowork trial with staff outside software development.
Vendor customer story (Anthropic deployment guide)
Pilot described, no figures reported
Team
Thomson Reuters teams outside software development
Workload
Automation and light prototyping on everyday documents and data
Product version
Claude Enterprise with Cowork

Run it beside another tool3

Brex offers Claude beside ChatGPT and Gemini and lets each employee choose.

Tiny's co-founder rates Claude best for writing and keeps ChatGPT and Gemini for other work.

A Meta product manager uses Projects in both assistants and opened a Claude project when Meta approached him about a role.

Evaluated it1

Ethan Mollick tested Claude Cowork against ChatGPT Work on the same seminar task, and Cowork asked before sending.

Adopted it3

HUB International began rolling Claude out to its 20,000-plus employees and reports 2.5 hours saved a week in early use.

Yale gave faculty and staff Claude Enterprise, with a separate HIPAA environment for patient data.

PostHog runs a weekly Cowork check across about fifty teams' plans, and finds Cowork still bad at most writing.

Reported by others3

Jamf's HR team turned its performance review spreadsheet into a Cowork skill without waiting on engineering.

Zapier's product marketing team drafts homepage messaging in Cowork from a skill that holds its positioning rules.

Thomson Reuters skeptics came around on Cowork after a couple of hours of real work.