For Anthropic · November 5, 2026
What adoption looks like in a large consulting practice, and what’s helping people get there.
Glenn Trepeta
With you today
glenn.trepeta@pwc.comlinkedin.com/in/glenntrepeta
M&A and enterprise transformation consultant by background. For the last two years, in a role dedicated full time to AI.
A front-row seat to enterprise AI adoption at scale, building and scaling it inside a large firm.
My practice is the first at the firm to roll Cowork out to every practitioner, with a dedicated rollout. Elsewhere, people can request access, but it isn’t pushed to them.
Part 1
Claude can do far more than most people ever ask of it.
The adoption gap · How the tools arrived
2024–25
e.g. ChatGPT, Copilot, Claude chat
May 2026
e.g. VS Code with GitHub Copilot; Claude Code by request
October 2026
The same idea in a friendlier app
The adoption gap · Rolled out like software
What a software rollout covers
Waterline: what a usage report can see
What a capability needs
The capability lives in people’s habits, not in their access.
The adoption gap · How people use it today
They take a first draft from AI, then do the rest by hand.
One chat per question, no folder, and nothing carried over to the next one.
They picture something that answers questions, not an agent that works in their files.
The adoption gap · It keeps moving
Just as people got comfortable with chatbots, the ground moved. Each new idea builds on the last, and most people are still on the first two.
Where most people got comfortableWhat arrived since
What I hear most“I don’t have time to keep up.”
Nobody needs all of it. People need the next step from where they are, and training that keeps going after the rollout.
Part 2
By how they use it, and by role. Each group gets stuck for a different reason.
Where people are · By how they use AI
→ The goal for a large firm: move people from groups 1 to 3 into group 4.
Where people are · By role
Know the business best and haven’t moved their work into AI.
Used to asking AI questions, but most aren’t power users, and they’re still learning the business.
Sit in between and rarely use AI, or use it carelessly.
Where people sit today. Illustrative, not measured.
Where people are · Each role up close
Longer at the firm, and know the business deepest
Where they get stuck
AI hasn’t clicked for them, and they don’t trust the output. They don’t yet see what sitting it out costs.
Competitors are moving and clients are asking about AI.
Newer to the work, and many used ChatGPT all through college
Where they get stuck
Used to asking it questions, not handing it real work, and still learning the work itself.
Their output looks confident and carries errors they can’t spot.
The large group in between
Where they get stuck
Many don’t use it. Those who do haven’t yet built the judgment it takes.
One poor result, and they stop trying.
Where people are · The biggest mistake
The biggest mistake is assuming people know something about AI. The real gap only shows when you get them talking or see their screen.
What we assumeWhat we find on their screen
So spell out every step: where to click, what it’s called and why it matters.
Part 3
Most of what holds people back has nothing to do with the model.
Where it breaks down · The part you already know
Work across your email, calendar and files in Microsoft 365
Not switched on
Open any folder you work in
Three approved folders only
Add skills and plugins as you need them
Each one waits on an approval process
Send it a task from your phone
Not switched on
Use new Claude features the day they’re released
Each one waits to be switched on
People judge the tool on the version they’re given. Switch every setting on and the gaps in habits and ways of working are still there.
Where it breaks down · Shared folders
Developers in Claude Codethe GitHub repo
Knowledge workers in Coworkthe engagement folder
README and CLAUDE.md
↔CLAUDE.md, read first by everyone
The code itself
↔The working files: models, decks, memos
Docs and specs
↔Context: the deal, stakeholders, workplan
Commit history
↔Session logs, meeting notes and recaps
Everyone clones the same repo
↔Everyone syncs the same SharePoint folder
Why one shared folder per engagement
The folder is the home base. Most people still work from one chat, or from a folder only they can see.
Where it breaks down · Team workspace plugin
The gap it fillsOut of the box, Cowork is built for one person: each person’s chats and memory stay on their own laptop, and nothing records what the team has done. Engagement teams keep asking for a shared workspace, so this plugin fills that gap.
What the plugin does
The team’s engagement folder
CLAUDE.mdread first, by everyonecontext/deal, stakeholders, workplansessions/a running log of every chatworking drafts/what Claude makesIs there a view where we can all see what everyone is doing in Claude?
What the team gets
Even with the plugin, it’s only as good as the people using it
There are good and bad users of AI, and most people don’t know that yet. A count of sessions logged can’t tell them apart.
Where it breaks down · The workspace
Where it breaks down · Working habits
Where it breaks down · Back and forth
There was no silver bullet before AI either. The same work just took longer.
The first prompt
Create a status deck for Friday.
Weekly Status Update[Project name] · Week of [date]Workstream 1On trackProgress continues as plannedWorkstream 2On trackKey activities underwayWorkstream 3At riskSome delays identifiedKey risksResource availabilityTimeline dependenciesNext stepsContinue executionFollow up on open itemsA generic draft
What’s missing is context: who it’s for, what to use, what good looks like. It isn’t a limit of the model.
A few follow-ups
It’s for the CFO. Lead with the run-rate. Match last week’s deck.
TSA run-rate is tracking to planRun-rate vs planOn planDay 1 items open2Go-live dateUnchangedRun-rate in line with the costing modelPayroll and TSA exit still openNo change to go-liveMonthly TSA run-rateJulAugSepOct2Close
The same back and forth you’d have with a teammate who missed the meeting.
The output
Ready to send
Steering committee · FridayDraft for discussionTSA run-rate is on plan. Two decisions this week keep go‑live on track.TSA run-rate vs planActualPlanOct at planJulAugSepOctDecisions needed1Approve TSA exit orderCFO · unblocks IT cutover2Confirm payroll cutoverHR lead · HR moves to greenWorkstreamsFinanceITHRLegalSource: models/tsa-costs.xlsx, workstream trackers3Often better than the version built by hand
And in less time than doing it by hand.
Where most people stopAfter the first prompt, they decide they can do it faster themselves and go back to doing it by hand.
Where it breaks down · Back and forth
Most people take the first output as it comes. It often clicks when they ask for something they didn’t think it could do.
The first output
Clean, correct and generic
The default look when nobody asks for more, and a sure sign no one iterated. Most people stop here.
One more prompt
Give me five versions of this slide, each a different way to show the dates. Make one a road with a sign I can click at each milestone.
What came back · five versions
Five ways to show the same dates, from one prompt
Click a version to see it. On the road, click a sign.
Where it breaks down · Ways of working
Standard work can be set up once. Work shaped by the client changes every time, so people have to stay hands-on and get comfortable iterating.
← Set inputs, a standard outputShaped by the client, meeting by meeting →
Most consulting work sits on the right. The skill people need most is working it through with Claude, turn by turn, not handing it off in one prompt and deciding the tool is no good.
Where it breaks down · The Cowork screen
Type / for skills
Why it causes so many problems
In chat, a project grouped chats, held instructions and built up memory. Working out of the team’s real folder is a different idea, and it never carried over.
A project is a layer just for you. The folder holds the real files.
Pick the synced engagement folder, and everyone works from the same files and the same CLAUDE.md.
Without a folder picked, Claude can’t see the team’s files, and most of what it can do stays out of reach.
Where it breaks down · Chasing the next tool
Many people miss when AI was a chat box: ask a question, get an answer, nothing new to learn.
New tool,
same habits
What happened this month
A small Microsoft Copilot Cowork pilot has email and calendar switched on, and people asked why the firm isn’t using it instead.
Claude has the same Microsoft 365 connector. It isn’t switched on at the firm yet.
What carries over to any tool
Part 4
Meeting people where they are, and showing them what it can do.
Closing the gap · What sticks
Most people think too small. It clicks when they see what Claude can do out of the box, and outputs well beyond a plain slide.
“It can do that?”
The most basic one. Once people see that Claude works from the folder they pick, setup, context and sharing start to make sense.
Seeing a power user’s work, like the films and the Lookbook, opens it up. Their own day-to-day use rarely shows what’s possible, so they don’t think bigger.
They’re used to plain PowerPoint. An HTML deck or artifact with live demos, video and parts they can click gets them thinking past the basic slide.
Impa replies in Teams with the likely fix and a one‑minute film.
Once going back and forth is a habit, they ask for more: five versions of a slide to choose from, then build out the one they like.
“This is real for me now”
In a proposal or a meeting, and there’s no good answer ready.
A competitor came in with AI built into how they would deliver the work.
Closing the gap · Work in progress
Closing the gap · Example films
Under four minutesOne idea each
Each film shows one idea step by step, in plain language, so anyone can use it right away.
Films
Closing the gap · A weekly show
New every FridayWeek 3October 306:57
A weekly show I make with one partner, among the firm’s most advanced AI users. Each episode takes one big, little idea: a small change in how you work with Claude that makes a big difference.
Episodes
Closing the gap · The Lookbook
Explainers, podcasts, troubleshooting and discussions. People search for the error they’re seeing and get the film that fixes it.
Open the Lookbook↗Closing the gap · Impa
It lives in a Teams chat. For the Cowork rollout, it answers practitioners’ questions with the likely fix and a Lookbook film.
How it works for the Cowork rollout
People get unstuck without waiting on me
Every question shows which film to make next
One template, many Impas
Q&A
Thank you for the time. Feel free to reach out anytime.
Emailglenn.trepeta@pwc.com
LinkedInlinkedin.com/in/glenntrepeta
Everything in the AI folder is fair game, other projects included. Claude takes the Ace deck's format and an Ace letter's wording, and won't say so.
Only Greenfield's files are in reach. Its brief, its rules and its own documents. Switch project, switch folder.
Every file in the folder you pick is fair game. Read the folder name under the message box before you type, every session.
Claude has access to everything in the folder you pick, and everyone with access to it sees what Claude saves there. Give each project team its own folder, so each team sees only its own documents.
Aurora’s SharePoint library. Only the three of you can open it.
A separate library. Nothing from Aurora ends up in here.
A personal folder in your OneDrive for notes, drafts and anything not tied to one team.
↪ Before you start fresh, capture what matters to a .md in the folder. The folder is the memory that carries across chats; the chat is just the working space.
The prompt is much bigger than what you type. Before you say a word, it already holds the folder’s CLAUDE.md, a short description of each skill you’ve added, and anything in memory.
Claude doesn’t automatically remember yesterday’s chat. What it can pick up is what’s in the folder: the CLAUDE.md, and any file you point it to. The folder is the memory, not the conversation.
Instructions, rules, standards. Loaded in full at the start of every session, for everyone who opens the folder.
Want something remembered next time? Say so: “remember this” or “save this to notes/lessons.md”. Next session, point Claude at the file. If it isn’t written down, don’t count on Claude having it next time.
CLAUDE.md looks like.project-aurora / CLAUDE.md
# Project Aurora — carve-out support # Context For anything about the deal, read project-context.md. For who's who and what they want, read stakeholders.md. For the workstream plan, read plan.md. # Where things go Put drafts in working/<my-name>/. Finished deliverables go in deliverables/. # Who am I I'm Glenn. Prefix files I create with "gt-".
“Tell Claude where to look and how you work — don't paste the deal in here.”
A line or two each. The detail itself lives in project-context.md, stakeholders.md and plan.md — those can be as long as they need to be, and that's where the who's-who goes.
If it’s context you’d otherwise explain again, ask Claude to write it down. Next time, you tell Claude where to find it.
For context you’d otherwise repeat: a decision, a document review, where things stand.
notes/tsa-decisions.md.notes/tsa-decisions.md, then draft the TSA schedule.Claude reads the file when you point it there. CLAUDE.md is the only file it reads on its own.
For a format or a set of steps you’ll use again: the format and the steps, not this deal’s facts.
A saved skill sits in Customize. Type / and its name to call it directly.
For work that a teammate, or a fresh chat, has to pick up.
handoff/tsa-status.md: where things stand, what’s open, which files matter.handoff/tsa-status.md and pick up from the open items.To check the note works, open a new chat, point it at the note and see whether it can carry on.
If you don't tell Claude, it doesn't know.
What most people type
Draft the steerco deck for Friday. Use this week's notes. Make it look good.
Said out loud, the way you'd tell a teammate
OK, I need the steerco deck for Friday. It's going to the seller's CFO. She reads the exec summary first, and what she actually cares about is the TSA run-rate, so lead with that. The point of the deck is to get a decision on the two open Day 1 items; it is not a status update. This week's call notes are in notes/, and the costing model in models/tsa-costs.xlsx is the number of record, so if the notes and the model disagree, the model wins. Match the last pack in templates/steerco-2026-08.pptx and keep it to five slides. One thing not to touch: the go-live date is agreed, don't reopen it. And if anything's unclear, flag it rather than guess.
Talk it through instead of typing it. It doesn't have to be tidy or in order: getting the context across matters more than a well-worded prompt.
If you don't tell it, it won't know.
A file, a chat message, a voice note. Any format works. The only way to get this wrong is not to do it.
The context window is everything Claude can see at once: your instructions, every file it has opened and its own replies. A data room is bigger than that.
When that doesn't happen, the conclusion is "the tool isn't worth it."
Some deliverables are one-shot, when the context is complete. Most need your judgment: that's the work, not a failure.
The failure hides inside good writing. Six ways it happens, all of them on real engagements.
The prose reads best where the understanding is worst.
Whatever it read feels like everything. Unswept isn't the same as nonexistent.
"All schedules reviewed" is generated text, not performed work.
Asked to shorten, it trims every section equally. The critical one loses as much as the boilerplate.
Twelve of forty schedules read — and nothing in the output tells you.
Legalese in, legalese out, whoever the audience is.
The counter is always something you can check: which documents it actually read, what it cut, a number re-derived from its source.
We hear it a lot. But the fire drill is the moment AI should help the most, as long as your workspace is set up and the draft was built there.
If the first draft that went to the partner was built in the folder, every time-sensitive ask after it is a one-sentence change to something that already exists.