The lines we hold
AI helps us go deeper and be better. Wisdom, judgment, discernment, connection, guidance, and acumen are ours. AI supports us in each of them. Draft wording until Dave Clapp approves the agreement
“We use AI to go deeper for you, not to cut corners.”
“Your information stays in our secure firm account, and it never trains anyone's AI.”
“Every judgment about a candidate, and every piece of advice, comes from a person who did the work.”
Our three boundaries
- AI never decides who advancesAI does not assess candidates. It helps us assess them. It never scores, ranks, or narrows a long list to a slate.
- AI never takes the place of a conversationIt does not interview, and it does not speak for us.
- The thinking is oursAI can help us gather, draft, and check. The questions, the read on a candidate, and the advice come from the person who did the work.
- Do all CarterBaldwin work in the firm's Claude account.
- Treat client and candidate information in Claude exactly as we do in LOXO.
- Read every AI output before we use, store, or send it, and check the facts.
- Write notes, summaries, and CardView entries from our own conversations.
- Ask Meredith Malec before connecting Claude to email, calendar, or any other system.
- Bring a skill to Tatum Preston before sharing it with the whole firm.
- Tell everyone on a call before any recording starts.
- Report a mistake the same day. Honest mistakes reported promptly are handled without blame.
- Ask when we are unsure.
- Let AI score, rank, or choose which candidates advance.
- Let AI interview, screen, or message a client or candidate on its own.
- Let an AI agent send, book, or pay for anything in the firm's name without a person's approval, or give any AI tool a firm credit card.
- Turn a transcript into notes or advice without doing the thinking ourselves.
- Put client or candidate information in a personal account or unapproved tool.
- Use a recording or transcription tool the firm has not approved.
- Enter Social Security numbers, account numbers, home addresses, or health information.
- Use age, gender, race, or any other protected characteristic in sourcing or assessment.
- Invent or embellish a candidate's credentials, or skip a reference or vetting step.
- Share anything that could identify a client, even without a name.
Everything else is open. Use AI however makes your work better.
Source: AI Use at a Glance, part of the agreement every employee signs.
Back to top ↑Your skills
A skill is our process, written down so Claude can follow it. Each one has one named owner. Find the skill for your role, type its name in a Claude chat, and attach your material. Confirm: typing /cb lists the firm skills for everyone
CB-Search-Launch
Drafts the kickoff set: milestones, position profile, call script, criteria.
CB-Research-Target-Guide
Builds out the research criteria and strategy that Search Launch drafted.
CB-Consultant-Interview-Coach
Checks your interview covers transitions, comp, alignment, motivation, personal.
CB-Milestones-Summarizer
Checks each meeting's goals, deliverables, and next steps against the Milestones.
Every output from a skill is a draft. You finish it.
Other skills in the library Confirm each row with Tatum
| Phase | Skill | What it does | Owner | Who has it |
|---|---|---|---|---|
| 1 · Launch | CB-Branded-Position-Profile | Turns the draft profile into the branded, client-ready version | Jodi Delikat | Firm-wide |
| 3 · Recruiting | CB-Consultant-Outreach | Drafts outreach in the consultant's own voice | Derek Broyard | Team |
| 5 · Slate to close | CB-Client-Interview-Guide-HP | One common interview frame for a client committee. It evaluates no one. | Bill Peterson | Team (HP) |
| 6 · Business development | CB-Brochure-Builder | Client brochures on the firm template | Allison Cohn | Firm-wide |
| 6 · Business development | CB-Outcomes-Writer | Search outcome announcements | Melissa Murray | Team (marketing) |
| 6 · Business development | CB-Slide-Visual-Updater | Updates slide visuals to the brand standard | Melissa Murray | Team (marketing) |
The six phases of a search: 1 Launch, 2 Research, 3 Recruiting, 4 Interviews, 5 Slate to close, 6 Business development.
Back to top ↑How the library runs
- One owner per skillThe owner writes it, updates it, and runs its review. Only the owner changes the instructions.
- CB- means firm skillOnly skills whose names start with CB- are firm skills. Firm-wide skills are already in your list. Team skills are shared with you by name.
- Use it as writtenChange the output in your own chat as much as you like. Do not edit the skill itself.
- Two checks on every skillA fact check reconciles numbers across transcript, resume, and job description. A bias check stops output arguing why a candidate fits instead of reporting what is true.
- Reviews on a calendarEach skill is reviewed 30 days after the October 13 launch, then every 60 days.
- Build your ownYou can build a skill for your own use. If it would help others, bring it to Tatum.
1. What worked. 2. The gap. 3. One change you would make.
Six daily moves
From the practice month. Copy the prompt and fill the brackets.
Attach the file
Attach the document itself. Copying text in loses the tables and layout.
Talk instead of type
Tap the microphone, talk for two minutes, then send this.
Think out loud
Say it in voice mode on the Claude phone app.
Give two examples
Use your own work. Add one you do not like and say why.
Ask for a table
Quicker to read and easier to check than paragraphs.
Start a new chat
Long chats give worse answers. Paste the summary into a fresh one.
Tools we built
Daily Practice
Short daily AI exercises. This is where the six moves above come from. New hires can start at day one.
cb-daily-practice.lovable.appExpedition Profile
The assessment we took in August and October to measure where the firm stands on AI.
expedition-profile.lovable.appWhat happened, by date
- AUG 3Kickoff in the Monday staff meetingThe baseline assessment opened the same morning.
- AUG 12Workflow working sessionWe mapped a search into six phases and listed the skills each one needs.
- AUG 17Current-state assessment shared with the firmAs reported that day, with 33 answers in: use of AI scored 79 out of 100, and sharing and learning from each other scored 61.
- AUG 20Ten staff interviews finishedSummary read out August 27.
- SEPT 1Daily Practice began
- SEPT 22First firm-wide skill publishedCB-Search-Launch, owned by Bill Peterson.
- OCT 8Second assessment opened18 people answered by October 12.
- OCT 13All-Staff AI DayThe skill library launched. Each group tested its skill, then every team fixed AI-drafted search documents with planted problems.
- OCT 28Research and admin team check-insCB-Research-Target-Guide had run 26 times in 30 days, by 8 people.
- NOV 2Consultant Interview Coach reviewConsultants agreed how to log intake notes, interview notes, and transcripts in Loxo, and to run the coach on every candidate bound for a partner.
- NOV 20Engagement closesTatum Preston owns the library from here.
The numbers
- 34 people took the August assessment
- 18 took it again in October
- 10 staff interviews
- 8 weeks of practice
The assessment scores four things from 0 to 100. Higher means more.
| What it measures | August 34 people | October 18 people | Change |
|---|---|---|---|
| How much we use AI | 77 | 84 | +7 |
| How comfortable we are with the tools | 63 | 74 | +11 |
| How much we expect AI to change our work | 72 | 72 | 0 |
| How much we share and learn from each other | 61 | 64 | +3 |
Read these with two cautions. Only 18 people answered in October, and they answered October 8 to 12, before the All-Staff AI Day. So the October column shows movement during the practice weeks. It does not measure the day. Fourteen people answered both times. For those fourteen the changes were smaller: use 79 to 83, comfort 68 to 72, outlook 69 to 68, sharing 63 to 65.
The August 17 report gave individual use as 79. One more person answered after that date, which moved the August average to 77.
A third round is not scheduled. Confirm with Dave and Tatum
Back to top ↑What we learned
- A polished draft can still be wrongWhat we saw On October 13 we worked on four AI-drafted search documents. Each carried 20 or more planted problems, and each read as finished.So what Before anything leaves the firm, ask “What did I add that the AI could not have?” and check each fact against the source.
- We built well alone and shared poorlyWhat we saw In August, use of AI scored 77 and sharing scored 61. Eleven skills were in use with no shared library, no named owner, and no version control.So what The library exists to close that gap. Every skill has one owner.
- Sharing is still the slowest thing to moveWhat we saw By October, sharing had moved from 61 to 64 while comfort with the tools moved from 63 to 74.So what When a skill works or falls short, send its owner the three things: what worked, the gap, one change.
- One version of each skillWhat we saw Before October 13, some of us had two copies of the same skill in our lists.So what The CB- copy is the only one. If you see two, tell Tatum.
- Chat count does not measure useWhat we saw Several of our heaviest users have few chats, because they work in Cowork and in files.So what Do not judge your own use, or anyone else's, by how many chats there are.
- Voice makes the hard exercises shorterWhat we saw The first practice exercise ran well past its planned fifteen minutes when done at a desk.So what Paste the prompt, turn on voice mode, and let Claude interview you on a walk or a drive.
What happens next
| When | What | Who |
|---|---|---|
| Nov 16 | Review of the agreement, We Strive for 100% Trust | Dave Clapp, Tatum Preston, Jason Malec |
| Nov 20 | The engagement closes. The skill library stays with its owner. | Tatum Preston |
| Every 60 days | Each skill is reviewed. Add the next review date for each skill | The skill's owner. Tatum keeps the calendar. |
| When someone joins | AI orientation, and a signature on the agreement | Name the owner |
| Every year | All-staff AI training, and a new signature on the agreement | Name the owner |
Documents
| Document | What it is | Link |
|---|---|---|
| How AI Supports Our Strategy | One page. How AI helps us go deeper and be better, and the three lines we hold. | Open |
| We Strive for 100% Trust | The confidentiality, AI use, and social media agreement. Signed at hire and every year. | Add when Dave approves |
| AI Use at a Glance | One page. The three words, the three boundaries, and the always and never lists. | Add when Dave approves |
| Claude cheat sheet | The one-page desk card from October 13. | Reprint first. It predates AI Use at a Glance |
| AI Skill Library | Every firm skill, its owner, and who has it. | Needs a current list. The Drive copy is from September 27 |
| All-Staff AI Day deck | The slides from October 13. | Open |
| Make It Great documents | The four AI-drafted search documents with planted problems, and the answer key. | Not in Drive. Tatum emailed them October 13 |
| Kickoff deck | The slides from the August 3 staff meeting. | Open |
| Kickoff leave-behind | The one-page roadmap handed out on August 3. | Open |
Keep learning
The AI Practice Guide is eight weekly habits, each with a prompt to copy. It is the eight-week practice we ran from August to October, rewritten so a new hire, or anyone who missed a week, can do it alone. Twenty to thirty minutes a week, on the search you are running.
- Weeks 1 to 4Give it more context. Talk first, then write. Show two examples. Force specifics.
- Weeks 5 and 6Ask it to find the hole in your own work. Verify what you did not write, and know when to stop.
- Weeks 7 and 8Write your prompt down with a named owner and a review date. Hand it to one colleague.
The guide is written for any organization. Where it and our agreement differ, the agreement wins.
Back to top ↑Test yourself
Eleven situations. For each one, choose Go for it, Never, or Ask first. Hold until after the November 16 staff meeting. Card 11 depends on the exceptions clause
- 01You use the firm's Claude account to turn your interview transcript into candidate notes.
- 02You paste a candidate's compensation and references into your personal ChatGPT.
- 03You record a candidate call with Plaud and don't mention it.
- 04You ask Claude to narrow 40 candidates down to the 6 you should advance.
- 05You use Perplexity to research an industry, with nothing confidential in the prompt.
- 06You want to connect a personal AI agent to your work email.
- 07An AI agent books a candidate's flight on the firm credit card.
- 08You build your own skill and want to share it with the firm.
- 09You have Claude draft a LinkedIn post about a placement that hasn't been announced.
- 10You realize you pasted a client's succession plan into the wrong tool.
- 11You think paid ChatGPT writes better and want to use it for firm work.
“Yes. It helps us go deeper for you. Your information stays protected. And every judgment is human.”
Your checklist
0 of 7 done Your ticks are saved in this browser only. No one else sees them.
Who to ask
Terms
- Skill
- Saved instructions that tell Claude how to do one job our way.
- Firm-wide skill
- A CB- skill everyone at the firm has. Only its owner changes it.
- Team skill
- A CB- skill shared with one named group.
- Personal skill
- A skill one person built for their own use.
- Connector
- A link that lets Claude read from a tool we use, such as SharePoint or Outlook.
- Cowork
- The mode where Claude works in your files and carries out several steps, instead of only chatting.
- Prompt
- What you type or say to Claude.
- Hallucination
- A statement from AI that sounds right and is false. The reason we check every fact against the source.