Partner Basecamp · Cohort 24

Function	Practitioner	Sr Practitioner	Mgr / Sr Mgr	Director+	Partner / MD	Total
Engineering	5	7	5	–	–	17
Architecture	1	4	3	5	–	13
Business Leadership	–	–	2	2	1	5
Project / Engmt	7	4	8	4	–	23

Segment	Exploring	Applying	Delivering	Frontier	n
Transformation Lead	39%	43%	11%	7%	28
Developer	18%	35%	35%	12%	17
Architect	23%	38%	23%	15%	13

Segment	Exploring	Applying	Delivering	Frontier	n
Manager or Senior Manager	33%	50%	11%	6%	18
Senior practitioner (5–9 years)	13%	13%	60%	13%	15
Practitioner (0-4 years in role)	23%	77%	—	—	13
Director, Senior Director, or Principal	55%	18%	—	27%	11
Partner, Managing Director, or Executive	—	—	100%	—	1

Segment	Too basic	About right	Too advanced	n
Transformation Lead	4%	71%	25%	28
Developer	18%	71%	12%	17
Architect	15%	85%	—	13

Segment	Too basic	About right	Too advanced	n
Applying in practice	4%	78%	17%	23
Learning and exploring	—	76%	24%	17
Delivering independently	33%	58%	8%	12
Operating at the frontier	17%	83%	—	6

Segment	Too slow	Well paced	Too fast	n
Transformation Lead	—	61%	39%	28
Developer	6%	82%	12%	17
Architect	—	69%	31%	13

Segment	Too slow	Well paced	Too fast	n
Applying in practice	—	78%	22%	23
Learning and exploring	—	41%	59%	17
Delivering independently	8%	83%	8%	12
Operating at the frontier	—	83%	17%	6

How likely are you to recommend attending this programme to a colleague?

n=64 Day 2 respondents · 110% of Day 1

Score distribution · 0 = not at all likely · 10 = extremely likely

Cohort	0	1	2	3	4	5	6	7	8	9	10
Cohort 24				1		2	3	9	14	15	20

■ Detractors 0–6■ Passives 7–8■ Promoters 9–10

NPS Breakdown

55%

36%

NPS +45 (n=64)

Promoters (9–10)Passives (7–8)Detractors (0–6)

Programme mean: +42 · 17 cohorts

Promoters mean: 51%

Passives mean: 39%

Detractors mean: 9%

NPS by Segment

Organisations, personas, and experience levels · ≥4 respondents shown

By Organisation

Promoters (9–10)Passives (7–8)Detractors (0–6)

DXC Technology

69%

31%

NPS +69 (n=13)

Deloitte

67%

33%

NPS +67 (n=6)

PwC

60%

40%

NPS +60 (n=5)

KPMG

60%

10%

30%

NPS +30 (n=10)

Fractional AI / Ode

80%

20%

NPS -20 (n=5)

By Persona

Promoters (9–10)Passives (7–8)Detractors (0–6)

Developer

69%

25%

NPS +62 (n=16)

Architect

60%

30%

10%

NPS +50 (n=10)

Transformation Lead

43%

48%

10%

NPS +33 (n=21)

Programme means · 17 cohorts: Architect +51 · Developer +47 · Transformation Lead +36

By AI Experience Level

Promoters (9–10)Passives (7–8)Detractors (0–6)

Operating at the frontier

60%

40%

NPS +60 (n=5)

Learning and exploring

60%

33%

NPS +53 (n=15)

Applying in practice

56%

33%

11%

NPS +44 (n=18)

Delivering independently

44%

11%

NPS +33 (n=9)

Programme means · 17 cohorts: Applying in practice +44 · Learning and exploring +44 · Delivering independently +42 · Operating at the frontier +37

What is the main reason for your score?

n=51 responses · organised by NPS segment

Promoters (score 9–10)· 27 responses

Great overview of key concepts

10“I thought that the basecamp was a really good way to get comfortable “looking under the hood” of agents and how to build them. Familiarizing concepts and bringing them to light, especially cost optimization. My answer to questions 5 & 6 is an improvement from a 2 heading into this.”

Passives (score 7–8)· 20 responses

It’s a good way to spend time trying the skills and great knowledge to take away.

8“I was a bit confused on the target audience for the breakout sections (technical vs non-technical). I think it was trying to walk a line between the 2. The “fill in the blank” activities were not that helpful. Would have been better to start with a problem and gradually build the solution thru multiple activities throughout the 2 days to layer on functionality and refining the process. The notebooks were very large, made it difficult to scroll back and forth and losing place in file.”

Detractors (score 0–6)· 4 responses

I think it has a complicated audience it’s targeted to. It needs to be designed more for either experienced or beginners. Right now it sits somewhere in the middle and disadvantages both groups. Also the course is too execution heavy and not enough on the conceptual side. Theres limited to learn from executing code and filling in blanks.

5“I think it has a complicated audience it’s targeted to. It needs to be designed more for either experienced or beginners. Right now it sits somewhere in the middle and disadvantages both groups. Also the course is too execution heavy and not enough on the conceptual side. Theres limited to learn from executing code and filling in blanks.”

1	2	3	4	5	Mean
2%	3%	16%	41%	39%	4.12

1	2	3	4	5	Mean
2%	2%	9%	50%	38%	4.20

1	2	3	4	5	Mean
—	3%	9%	50%	38%	4.22

NPS Reason

n=51 responses · grouped by NPS segment

Promotersscore 9–10 · 27 responses

1Great overview of key concepts10

2This was an awesome learning experience10

3Hands on time10

4Foundational knowledge about how to think about using agents.10

5Great learning10

6It was great starting step to know the Claude10

7The course increases comfort with AI while demonstrating abilities and uses that are key for client adoption.10

8Good balance of theory and hands-on10

9I learned a lot! It was fun!10

10The buildup of the workshops to the final hackathon that encompasses it all. The laid-back nature of the workshops. It felt more like a learning space rather than anthropic sale pitch.10

11Provides an excellent coverage of what Claude is and can do. And efficiencies can be built.10

12I thought that the basecamp was a really good way to get comfortable “looking under the hood” of agents and how to build them. Familiarizing concepts and bringing them to light, especially cost optimization. My answer to questions 5 & 6 is an improvement from a 2 heading into this.10

13The session was clear, practical, and included real-world examples that I can apply in my work.10

14Everything was great10

15Well structured contents.. already thinking of applying some in the day-to-day work10

16Great bootcamp, great learnings and organization10

17Excellent coverage of core concepts10

18Super helpful course and really allowed me as a beginner to put into practice.9

19Satisfied9

20It was a well organized session. Lots of good learning9

21Great insights of AI and hands-on activities.9

22I love it , I had great experience , networking, I would have liked if we had more time and the api calls keep timing out or hang in transit . As a developer and architect fast iteration cycle is important to create solutions fast9

23Very detailed and liked the hands on experience9

24It provides good over view of all concepts with some hands on. It will be great if some of these exercises can be added as prep work on skill jar.9

25I learned a lot!! It’s great to be surrounded with people all trying to learn how to use Claude better and build.9

26Very informative, very hands-on, and surprisingly fun for a corporate training. It is a fantastic workshop for anyone knee-deep in the AI wave9

27Trust Claude expertise and accelerate the product delivery with lower cost9

Passivesscore 7–8 · 20 responses

1Good content8

2Good training for non experts8

3Deff new learning on AI capabilities8

4Very hands on program, with a good cross section of examples to work through8

5The ability to ask Anthropic engineers actual questions8

6PwC should have allowed us to do it on our work laptops in order to attend this meeting. Would have been a lot more helpful.8

7Beneficial however is very technical and may be difficult for those with little technical experience8

8It's very good and lot of insights but if the program start with basic to intermediate and then to building agents it would be more useful8

9Great deep dive into Claude code to answer WHY and HOW Claude is building rather than just accepting the outputs of claude8

10Lots of good technical grounding and use cases that resonated. Moves fast and might have needed a bit more technical background/ instruction. Very good facilitation.8

11I would like my non-engineer colleagues to take courses and join this base camp session for them to know what is happening behind the scene of Claude and being able to more compellingly propose ai/claude to clients.8

12I was a bit confused on the target audience for the breakout sections (technical vs non-technical). I think it was trying to walk a line between the 2. The “fill in the blank” activities were not that helpful. Would have been better to start with a problem and gradually build the solution thru multiple activities throughout the 2 days to layer on functionality and refining the process. The notebooks were very large, made it difficult to scroll back and forth and losing place in file.8

13It’s a good way to spend time trying the skills and great knowledge to take away.7

14great to get a basic understanding of ai7

15The training was a good basis but with time permitting I’d rather build from the start rather than running premade7

16This is for the beginner level folks7

17There wasn’t enough time to complete the exercises, but overall helpful7

18I think it is helpful but for an engineer already using AI there’s a lot of repeat. I think it would be helpful to lean into a engineering or non-engineering boot camp.7

19Could use some improvements in the exercises, eg should not use notebooks. The technical levels appropriate for product and newer AI engineers7

20At times I didn't fully understand what was going on. Trying to troubleshoot problems while working on hands on work. The hackathon piece was the most interesting and it would be great to see if you could incorporate that with the hands on work we did7

Detractorsscore 0–6 · 4 responses

1It’s too fast , material can be presented in a more understandable manner6

2I’m not sure if it is designed for engineers or non-engineers6

3I think it has a complicated audience it’s targeted to. It needs to be designed more for either experienced or beginners. Right now it sits somewhere in the middle and disadvantages both groups. Also the course is too execution heavy and not enough on the conceptual side. Theres limited to learn from executing code and filling in blanks.5

4Most of my colleagues build agentic software and while it was useful, most of this material has already been covered in the various certs. So I think this needs to be clear it’s not as much targeted to people who are very far into the anthropic product space., the hackathon was fun and the hosts were educated and helpful5

Most Valuable

What one takeaway will you share with a colleague or client? · n=43 responses

1Do the work later in the night when you have more time.

2It's not the model

3Build with vision to prod

4It’s more about the process than the model capability

5The different use cases for the different Claude models

6Token usage can be reduced

7How collaboration and sharing insights / learnings is still a critical component of our daily lives.

8Costing methods to decrease expense

9Watch the change and adapt the change

10eval structure

11Tokenomics/ Cost optimization techniques

12Claude AI next Gen

13That Claude is easy to use. You just need to know the pitfalls of utilizing AI for coding

14The managed agents paradigm is something I’ll definitely keep in mind in client conversations moving forward

15Play with the effort level in evals

16How to optimize agentic harness and architecture

17Start small with building agents and then iterate

18You need to learn by doing

19It is almost never the model it is the code around the model

20Architecture of your agentic solution is super important, i learned here about how to optimize

21Compact lets you describe how the compaction occurs

22Building agents is not as difficult or scary as it may seem starting out.

23Evaluating before implementing is the key part in AI journey for an enterprise.

24The session is interactive and lot more to explore

25Having a subagent reach out when it wants help

26Always be open to learn. Be a part of the change or it will leave you behind.

27Context matters a lot, and you want to give the right amount of context and engineer it optimally

28Efficiency.

29The newest model is not always the best model to use for the use case. They should be fit for purpose and have controls in place to improve accuracy rather than choosing a more expensive and newer model by default

30Eval, cost optimization, context engineering

31Context engineering. Didn’t know this was so important!

32How great the food was. On a serious note, the different ways we can push back on cost questions and helping clients understand that Opus is not always the answer but there’s smaller things we can incrementally before we arrive at that conclusion.

33That you can use # to edit the memory

34Try claude you will not regret it

35AI is an extension of human intelligence. It’s good enough to match us, if we are thoughtful in our design. It’s not perfect. It requires us to re-think everything.

36It is not always the model

37Diagnostic, context engg and Eval

38The cutting-edge AI products that actually work and make money are much more than just throwing in a model and a prompt. AI, especially Claude, is nowadays far more capable and has many more additions and features to optimize complex use cases

39The fact that not always the cheaper model will lead to the cheaper solution

40Claude is the most trusted and efficient AI tool out there

41Learnings around context engineering and model attributes

42AI is not the matter within our sector. Let’s see more about what happening in the world.

43Need to be prepared to run python notebooks. Might be helpful to have a couple helpful prework to make sure environment is setup before hand

30-Day Intentions

What will you build for yourself or apply at work? · n=44 responses

1A new game with AI agent features

2Code analyzer

3Workout agent

4I’ll continue playing with building apps

5Build new apps

6An agent to help evaluate my individual finances.

7I build a lot of demo and proof of concept

8The hackathon product we produced has real world potential and we will pursue it.

9Multi agent work

10Eval and inference optimization

11An Agent that can help in my Data Assessment

12Few agents

13I will build out tools to better help my build Workday integrations

14Some of the long term memory management concepts from the hackathon I’ll definitely try to incorporate

15I’ll advise clients slightly differently

16Will go back through the examples we used plus build a cross section of agents next that build on the concepts we used in the camp

17A data analytics platform for a large bank

18I am building a ai development course for my team that has agents as teachers.

19Extend on my hackathon project

20Build an agent for my current project to do my task, host all the knowledge, etc

21Sub agents, MCP, team agent

22Client’s project is building an agentic system

23Agents that can help optimize the order to cash processes

24Identify workshops that can be agentified.

25Try to build an agent to crawl through different applications and provide error free code

26Too many things to namr

27Experiment , test it and develop and improve it, and tell others how

28A multi agent claude model agenta fpr doing my various day to day task.

29I am currently trying to brainstorm an agentic flow for my client, who has never used AI tools before and has a ton of manual processes

30Develop at least 2 work relevant applications

31Deep review of the code that we have had Claude build and what mechanisms are in place to reduce cost and improve accuracy based on the levers that we discussed in class

32Increase Claude adoption for developers.

33Continuing building our all we started here at base camp.

34Custom chat bot for support with TMS implementations.

35Intelligence per task optimization and skill prompt minimization

36I'm building an AI agent that acts as a copy of myself, taking over some of my routine responsibilities so I can focus on higher-value work.

37Better evals

38On Eval

39The use of better context engineering, leveraging and optimizing system prompts, model settings/configs to optimize for latency, response quality, cost, etc in my current work projects.

40Review the learnings and try to apply them live

41MCP integrations, and Agents

42Will apply learnings to our public cloud practices

43I will try building demo agent to be added to my proposal deck. And share the experience and enthusiasm I got in Basecamp to my colleagues in Japan.

44Apps with Agents.Build MCP to use with applications

Day 1 responses	58
Day 2 responses	64 (110% of Day 1)
Matched pairs	47 (73% of Day 2)
Orgs resolved	58 of 58 respondents matched to named org
Day 2-only respondents	17 (26.6%) — suppresses NPS by 2 points; no Day 1 data, excluded from persona & confidence analysis

San Francisco