← Home
Partner Basecamp · Cohort 33

San Francisco

September 17–18, 2026
+40
Net Promoter Score
67
Day 1 Responses
86%
Matched Pairs
92%
Day 2 Response Rate
Programme Satisfaction
Positive NPS of +40 — promoters outweigh detractors with moderate passive presence.
+40NPS · n=62
50% Promoters40% Passives10% Detractors
95% CI: +24 → +57  ·  True NPS lies within this range with 95% confidence (n=62 respondents)
Confidence Arc
Strong end-of-Day-1 build confidence — cohort leaves Day 1 ready to apply Claude in client work (mean 4.1/5).
4.1end-of-D1 build / 5
4.2D2 Design / 5
4.3D2 Commercial / 5
4.3D2 Build / 5
Audience
67 participants across 16 organisations — Transformation Lead majority with applying in practice the most common AI experience level.
Top Organisations
LTM (13) PwC (9) Deloitte (8) Altimetrik (7) Ascendion (5)
Voice of Participant
What participants said — dominant theme from each NPS segment.
Promoters
Very detailed thorough presentation and exercises
Passives
Get insights on Claude concepts
Detractors
No advanced concepts covered; differentiation with other frontier models from different organizations not explained. The clients want to know “why Claude or Anthropic”. Don’t have an answer to that. Basic concepts covered; nothing much in scope for an AI architect already working in the field.. maybe the course is designed like this but proper expectations should be set to allow the partner organizations to identify the right candidate/cohort.
Survey & Data Quality
Survey & Data Quality
Day 1 responses67
Day 2 responses62 (92% of Day 1)
Matched pairs53 (86% of Day 2)
Orgs resolved67 of 67 respondents matched to named org
Day 2-only respondents9 (14.5%) — suppresses NPS by 2 points; no Day 1 data, excluded from persona & confidence analysis
Organisations
All Day 1 respondents · n=67
LTM13 (19%)
PwC9 (13%)
Deloitte8 (12%)
Altimetrik7 (10%)
Ascendion5 (7%)
NEC5 (7%)
Grant Thornton4 (6%)
Capgemini3 (4%)
IBM3 (4%)
McKinsey2 (3%)
Blank Metal2 (3%)
Lovelytics2 (3%)
Accenture1 (1%)
Cognizant1 (1%)
Wipro1 (1%)
AWS1 (1%)
Function × Seniority
All Day 1 respondents · n=67
FunctionPractitionerSr PractitionerMgr / Sr MgrDirector+Total
Engineering584421
Architecture254718
Business Leadership––145
Project / Engmt4310623
Experience Profile
AI experience level · n=67 · % of row
By Persona
SegmentExploringApplyingDeliveringFrontiern
Transformation Lead25%61%14%—28
Developer24%57%10%10%21
Architect6%56%22%17%18
By Seniority
SegmentExploringApplyingDeliveringFrontiern
Director, Senior Director, or Principal24%52%10%14%21
Manager or Senior Manager11%68%21%—19
Senior practitioner (5–9 years)25%56%12%6%16
Practitioner (0-4 years in role)18%55%18%9%11
How would you describe your current experience with AI tools prior to Basecamp?
Learning and exploring
13 (19%)
Applying in practice
39 (58%)
Delivering independently
10 (15%)
Operating at the frontier
5 (7%)
Prior to today, how much had you worked with Claude or the Anthropic API?
Not at all
2 (3%)
A little
33 (49%)
Regularly
32 (48%)
Did the depth land for this audience?
Technical depth perception · n=67 respondents · % of row
By Persona
SegmentToo basicAbout rightToo advancedn
Transformation Lead14%64%21%28
Developer—86%14%21
Architect11%89%—18
By AI Experience Level
SegmentToo basicAbout rightToo advancedn
Applying in practice5%82%13%39
Learning and exploring—69%31%13
Delivering independently30%70%—10
Operating at the frontier20%80%—5
Overall depth distribution
Too basic
6 (9%)
About right
52 (78%)
Too advanced
9 (13%)
Did the pace work across the room?
Session pace perception · n=67 respondents · % of row
By Persona
SegmentToo slowWell pacedToo fastn
Transformation Lead—89%11%28
Developer—86%14%21
Architect—83%17%18
By AI Experience Level
SegmentToo slowWell pacedToo fastn
Applying in practice—87%13%39
Learning and exploring—92%8%13
Delivering independently—80%20%10
Operating at the frontier—80%20%5
Overall pace distribution
Well paced
58 (87%)
Too fast — not enough time to apply
9 (13%)
How confident were participants to build with Claude after Day 1?
End-of-Day-1 confidence · "How confident are you in your ability to build a client solution using Claude?" · n=67 · 1–5
Mean end-of-D1 build confidence by Persona
Architect4.50/5 · n=18
Developer4.24/5 · n=21
Transformation Lead3.82/5 · n=28
Programme mean: 3.9/5 · 24 cohorts
Mean end-of-D1 build confidence by AI Experience Level
Operating at the frontier4.80/5 · n=5
Delivering independently4.50/5 · n=10
Learning and exploring4.15/5 · n=13
Applying in practice3.95/5 · n=39
Overall end-of-D1 build confidence distribution
1
1 (1%)
2
2 (3%)
3
11 (16%)
4
26 (39%)
5
27 (40%)
How relevant was today's content to your current role?
Content relevance rating · n=67 respondents · 1 = not relevant, 5 = highly relevant
Mean relevance by Persona
Architect4.33/5 · n=18
Developer4.29/5 · n=21
Transformation Lead3.89/5 · n=28
Mean relevance by AI Experience Level
Learning and exploring4.31/5 · n=13
Delivering independently4.30/5 · n=10
Applying in practice4.08/5 · n=39
Operating at the frontier3.80/5 · n=5
Overall relevance distribution
1
0 (0%)
2
0 (0%)
3
17 (25%)
4
24 (36%)
5
26 (39%)
How likely are you to recommend attending this programme to a colleague?
n=62 Day 2 respondents · 92% of Day 1
Score distribution · 0 = not at all likely · 10 = extremely likely
Cohort012345678910
Cohort 331122520823
■ Detractors 0–6■ Passives 7–8■ Promoters 9–10
NPS Breakdown
50%
40%
10%
NPS +40 (n=62)
Promoters (9–10)Passives (7–8)Detractors (0–6)
Programme mean: +42 · 24 cohorts
Promoters mean: 52%
Passives mean: 38%
Detractors mean: 10%
NPS by Segment
Organisations, personas, and experience levels · ≥4 respondents shown
By Organisation
Promoters (9–10)Passives (7–8)Detractors (0–6)
LTM
77%
23%
0%
NPS +77 (n=13)
NEC
60%
40%
0%
NPS +60 (n=5)
Altimetrik
50%
50%
0%
NPS +50 (n=4)
Deloitte
50%
50%
0%
NPS +50 (n=4)
PwC
22%
67%
11%
NPS +11 (n=9)
Grant Thornton
0%
50%
50%
NPS -50 (n=4)
By Persona
Promoters (9–10)Passives (7–8)Detractors (0–6)
Developer
59%
35%
6%
NPS +53 (n=17)
Architect
50%
44%
6%
NPS +44 (n=16)
Transformation Lead
40%
50%
10%
NPS +30 (n=20)
Programme means · 24 cohorts: Architect +50 · Developer +50 · Transformation Lead +34
By AI Experience Level
Promoters (9–10)Passives (7–8)Detractors (0–6)
Learning and exploring
54%
46%
0%
NPS +55 (n=11)
Applying in practice
50%
43%
7%
NPS +43 (n=30)
Delivering independently
50%
38%
12%
NPS +38 (n=8)
Operating at the frontier
25%
50%
25%
NPS 0 (n=4)
Programme means · 24 cohorts: Delivering independently +47 · Applying in practice +45 · Learning and exploring +42 · Operating at the frontier +36
What is the main reason for your score?
n=43 responses · organised by NPS segment
Promoters (score 9–10)· 21 responses
Very detailed thorough presentation and exercises
9“Very good ans informative sessions. The hands on exercises are well crafted. The concepts covered were good and gives confidence to have claude conversations with partners.”
Passives (score 7–8)· 16 responses
Get insights on Claude concepts
8“Would advise a deeper familiarity with foundations and hands on experience prior to attendance. Else the exercises will take longer to properly explore than the session times provide. Good to have as takehome exercises though.”
Detractors (score 0–6)· 6 responses
No advanced concepts covered; differentiation with other frontier models from different organizations not explained. The clients want to know “why Claude or Anthropic”. Don’t have an answer to that. Basic concepts covered; nothing much in scope for an AI architect already working in the field.. maybe the course is designed like this but proper expectations should be set to allow the partner organizations to identify the right candidate/cohort.
2“I think the trainings should be a little bit more guided vs just having people gather in a room a say he is the zip file and go figure it out. It also didn’t help that PwC was constantly having access problems which is PwC’s fault for not helping us prepare ahead of time. Patricio was tremendous in helping us. I would say the most fun was the hackathon. But the main thing is the trainings should be more guided, because the way it was ran, that could probably be done online. If it’s in person, it should be more interactive with the instructors.”
End-of-Programme Confidence
n=62 respondents · 1 = not confident, 5 = very confident
How confident are you in your ability to design an evaluation for an AI solution?
12345Mean
—3%13%44%40%4.21
Programme mean: 4.1/5 · 24 cohorts
Leaving today, how comfortable are you advising clients on AI when questions arise?
12345Mean
—3%11%42%44%4.26
Programme mean: 4.2/5 · 24 cohorts
How confident are you in your ability to lead a conversation about Anthropic and Claude with a client?
12345Mean
—2%15%37%47%4.29
Programme mean: 4.2/5 · 24 cohorts
Confidence Trajectory by Persona
D1→D2 delta · apply-AI confidence · matched pairs only · n=53
Positive delta = confidence grew · negative = dropped
Architect (n=16)
+0.00
Developer (n=17)
+0.35
Transformation Lead (n=20)
+0.30
What one takeaway will you share with a colleague or client?
n=41 responses
Claude / Anthropic content
“AI is not just prompting .”
AI agents & engineering
“Advisory agent”
Evals & testing
“The need to understand and test the whole architecture to confirm it is working efficiently”
Tool setup / readiness
“The amount the non-model elements matter. Getting higher performance from cheaper models with better tooling around it.”
AI strategy & use cases
“The best model is not needed for most use cases. Use Sonnet”
What will you build for yourself or apply at work in the next 30 days?
n=40 responses
AI agents & engineering
“Rebuild some Claude agents that I use internally with more caching & memory”
Claude / Anthropic content
“Rebuild some Claude agents that I use internally with more caching & memory”
Evals & testing
“Double down on establishing quality evals”
AI strategy & use cases
“Agentic AI implementation for few use case around finance and supply chain”
Speed / pacing concerns
“Multiple agents in my work space”
Partner Basecamp Survey Report · Cohort 33 · Generated by analyze.py v2.5
NPS Reason
n=43 responses · grouped by NPS segment
Promotersscore 9–10 · 21 responses
1Very detailed thorough presentation and exercises10
2Hands on. - build along exercises10
3Learning10
4Interactive. Great material10
5Prompt test cases.10
62 days deliberate work, answers many questions for self and customer10
7Complete key walk thru on Claude code usage10
8Very informative and with real demo scenarios that are excellent for learning concepts10
9Colleague will learn the concepts and run practical scenarios10
10Topics covered in the base camp.10
11Learned a lot of things that you read on the internet but never actually got to work on. The hackathon at the end rounded everything we learnt over the span of 2 days.10
12Learned the holistic view and applied ai techniques10
13Teaches aspects of AI implementation that developers ignorantly ignore10
14It helped to understand the details how to optimize the agent development and reduce the cost for customer10
15I now understand the key points of how to embed AI into a system.10
16Informative10
17This was a very informative session9
18Very good ans informative sessions. The hands on exercises are well crafted. The concepts covered were good and gives confidence to have claude conversations with partners.9
19Important topics covered9
20Topics covered in the course9
21Learn about LLM models, evaluation, and observability.9
Passivesscore 7–8 · 16 responses
1Very interactive8
2Get insights on Claude concepts8
3Good knowledge8
4Corporate firewall/connectivity issues8
5Good for those familiar with coding to a certain extent8
6Interesting learning8
7It depends on the colleague8
8私はこの回答をclaudeが翻訳してくれると信じて書きます。私は、AIがすべきこと、人がやるべきことをいつも悩んでいました。この機会は，私にその答えをくれました。8
9It is very technical, so a manager may have a hard time with the hands on practice. I think the forcing seats with balanced skills is worth looking into.8
10Would advise a deeper familiarity with foundations and hands on experience prior to attendance. Else the exercises will take longer to properly explore than the session times provide. Good to have as takehome exercises though.8
11It's good for me that there are many hands-ons and I feel that I can obtain various skills about Claude.8
12The session covers all the basics and to some extent some advanced concepts as well.8
13Hands on was great but more guided hands on would have made it more meaningful7
14It is not for everyone. It would be a great addition to folks with some basic development background7
15I think the exercises are cool but I don’t think I really learned much in terms of capabilities7
16Overall a great program with great exposure. However it depends what you’re looking to gain out of it. Very technical rather than conversations about strategy and different successful use cases.7
Detractorsscore 0–6 · 6 responses
1Good6
2The content was really good, but I wish that we went through the actual “answer keys” at the end. To show exactly what an optimal solution for these exercises are.6
3No advanced concepts covered; differentiation with other frontier models from different organizations not explained. The clients want to know “why Claude or Anthropic”. Don’t have an answer to that. Basic concepts covered; nothing much in scope for an AI architect already working in the field.. maybe the course is designed like this but proper expectations should be set to allow the partner organizations to identify the right candidate/cohort.5
4Dev oriented, deep on solutions, issues and architecture. Very specific skill set is the target5
5PowerPoint not shared. Nothing novel. Code notebooks were sloppy.and unrefined. Not enough time for the hackathon.3
6I think the trainings should be a little bit more guided vs just having people gather in a room a say he is the zip file and go figure it out. It also didn’t help that PwC was constantly having access problems which is PwC’s fault for not helping us prepare ahead of time. Patricio was tremendous in helping us. I would say the most fun was the hackathon. But the main thing is the trainings should be more guided, because the way it was ran, that could probably be done online. If it’s in person, it should be more interactive with the instructors.2
Most Valuable
What one takeaway will you share with a colleague or client? · n=41 responses
1Caching
2Using ai to automate work
3AI is not just prompting .
4Advisory agent
5The amount the non-model elements matter. Getting higher performance from cheaper models with better tooling around it.
6The need to understand and test the whole architecture to confirm it is working efficiently
7There’s always room for improvement and to make it more effective
8The best model is not needed for most use cases. Use Sonnet
9Evaluation techniques
10Screenshots
11How the prompt is laid out is key
12The importance of context engineering… more is not always better
13It’s all under control with proper use of the technology
14Same as above
15Refine prompt, evaluate agents and choose the right model for the task
16Start using Claude code
17That model handoffs are great for me deference and cost optimization
18Its new way of delivering value, need to embrace and tune up to expectations.
19Wonderful knowledge experience on Claude
20The lesson that there’s a ton of efficiencies that can driven by a better agent setup that can make this cost effective.
21It’s never too late for anything in terms on learning and leaning on new technologies
22The model is rarely the answer.
23How to optimize the agents with various techniques
24More rigour about model selection is smart. Haiku is actually pretty capable
25評価の大切さ、モデルだけでは解決することができないことがあること。
26There are optimizing techniques that you can use to optimize how to efficiently write a workflow. It’s also a growing for field. Refresh your knowledge often.
27Switching to the most modern model requires more evaluation than it's one line switch would indicate. Must be tested
28Claude does make everyones life easier.
29AI is a powerful tool when you know when and how to use it.
30Useful AI agent needs more than a strong prompt, it needs clear instruction realistic evils and a way to diagnose failures
31The model is never the problem, the way you use it is
32Eval, Agents and mcps
33Agent can really bring the transformation
34How to implement it right
35Evaluation capabilities
36Initially I was thinking like if I am not getting good response switch the model but that was not the case always. Your prompt is always important
37How to optimize the agent development
38The most expensive model is not always the perfect solution for what you need
39I learned so much. Thank you for such a wonderful workshop!
40Bring your personal computer if you’re in pwc
41The importance of evals
30-Day Intentions
What will you build for yourself or apply at work? · n=40 responses
1Rebuild some Claude agents that I use internally with more caching & memory
2Agent to make my life easy
3Enterprise agents
4Personal finance assistant
5Double down on establishing quality evals
6Harnesses for evaluation of agents
7Multiple agents in my work space
8Too long to explain
9Improvements in AI agents I’ve been building.
10Agents to help with project management
11Multi project management dashboard, plug and play AI system evaluation harness
12Solution design patterns
13LLM doesn’t matter as much as everything else
14Create an agent to organize my work, to do some pre work before I actually log on to start my day
15We are looking at build a swarm of agents to improve operational excellence for data platforms
16Multiple agents
17Probably evals
18Agentic AI implementation for few use case around finance and supply chain
19Swarm agent for our needs
20I have a personal project building a budgeting app. I’ll apply these lessons to configuring the repo to make the most of Claude code.
21Something which is useful to my clients in my day-to-day job
22Agents!
23Project centralized and agent driven system
24Building Agentic transformation solutions
25Probably eval harnesses
26私が作った過去のAIエージェントに評価機構をいれます。
27Whatever idea that comes to mind.
28Include claude as a partner for every work
29I want to work and refine the hackathon project (team Starpeople) and also apply learnings to my work, specifically building and running agents to automate the repeated tasks
30Agents and mcps
31Every client opportunity with show and tell with a lead fir action mindset for building ai agents.
32An evaluated right prompted parallel token optimized agent working with suitable agent.
331. Use Claude to reverse engineer legacy applications Technical and Functional specifications 2. Usage of playwright mcp to build validation scripts of new changes in the applications.
34Apply Claude in everyday work and keep myself very productive.
35Build the agents with optimized architecture
36Evaluations for agents. TTFT and TTC monitoring.
37Demo app proving our ai-ready data concept
38How to improve client agents, low cost optimization, model optimization,
39Consider model-related issues and system-related issues separately.
40When things don't go well, don't blame the LLM model first.