Partner Basecamp · Cohort 27

Programme Satisfaction

Positive NPS of +43 — promoters outweigh detractors with moderate passive presence.

+43NPS · n=56

50% Promoters43% Passives7% Detractors

95% CI: +26 → +59 · True NPS lies within this range with 95% confidence (n=56 respondents)

Confidence Arc

Moderate end-of-Day-1 build confidence — cohort is progressing; Day 2 has room to close remaining gaps (mean 3.9/5).

3.9end-of-D1 build / 5

4.0D2 Design / 5

4.0D2 Commercial / 5

3.9D2 Build / 5

Audience

72 participants across 17 organisations — Developer majority with applying in practice the most common AI experience level.

Top Organisations

Capgemini (27) MHP (5) Deloitte (4) DXC Technology (4) Slalom (4)

Voice of Participant

What participants said — dominant theme from each NPS segment.

Promoters

Easy to follow sessions

Passives

Great information, very fast paced

Detractors

Not technical enough for my team

Segment Finding

Capgemini's NPS (+65, n=20, 0 detractors) was driven by a genuinely foundational-level room - 41% early/foundational AI experience. Promoters cite the hands-on format as what landed for them. Mirror image of cohorts where advanced practitioners find material too basic.

Analyst Note

Passive rate ran +4pts above programme average (43%). No single driver - mix of advanced practitioners wanting more depth, some feeling hands-on time was rushed, and light corporate-environment friction. Not a satisfaction problem: 0% detractors cohort-wide.

Function	Practitioner	Sr Practitioner	Mgr / Sr Mgr	Director+	Partner / MD	Total
Engineering	6	18	5	3	–	32
Architecture	3	6	7	5	–	21
Business Leadership	–	1	3	2	2	8
Project / Engmt	5	1	2	3	–	11

Function

Practitioner

Sr Practitioner

Mgr / Sr Mgr

Director+

Partner / MD

Total

Engineering

–

Architecture

–

Business Leadership

–

Project / Engmt

–

Segment	Exploring	Applying	Delivering	Frontier	n
Developer	19%	44%	25%	12%	32
Architect	24%	19%	19%	38%	21
Transformation Lead	47%	16%	11%	26%	19

Segment

Exploring

Applying

Delivering

Frontier

Developer

19%

44%

25%

12%

Architect

24%

19%

38%

Transformation Lead

47%

16%

11%

26%

Segment	Exploring	Applying	Delivering	Frontier	n
Senior practitioner (5–9 years)	19%	42%	27%	12%	26
Manager or Senior Manager	24%	29%	12%	35%	17
Practitioner (0-4 years in role)	64%	29%	7%	—	14
Director, Senior Director, or Principal	15%	8%	15%	62%	13
Partner, Managing Director, or Executive	—	—	100%	—	2

Segment

Exploring

Applying

Delivering

Frontier

Senior practitioner (5–9 years)

19%

42%

27%

12%

Manager or Senior Manager

24%

29%

12%

35%

Practitioner (0-4 years in role)

64%

29%

—

Director, Senior Director, or Principal

15%

62%

Partner, Managing Director, or Executive

—

100%

—

Segment	Too basic	About right	Too advanced	n
Developer	6%	94%	—	32
Architect	43%	52%	5%	21
Transformation Lead	—	74%	26%	19

Segment

Too basic

About right

Too advanced

Developer

94%

—

Architect

43%

52%

Transformation Lead

—

74%

26%

Segment	Too basic	About right	Too advanced	n
Applying in practice	—	95%	5%	21
Learning and exploring	5%	70%	25%	20
Operating at the frontier	41%	59%	—	17
Delivering independently	21%	79%	—	14

Segment

Too basic

About right

Too advanced

Applying in practice

—

95%

Learning and exploring

70%

25%

Operating at the frontier

41%

59%

—

Delivering independently

21%

79%

—

Segment	Too slow	Well paced	Too fast	n
Developer	—	94%	6%	32
Architect	10%	90%	—	21
Transformation Lead	5%	79%	16%	19

Segment

Too slow

Well paced

Too fast

Developer

—

94%

Architect

10%

90%

—

Transformation Lead

79%

16%

Segment	Too slow	Well paced	Too fast	n
Applying in practice	5%	95%	—	21
Learning and exploring	5%	75%	20%	20
Operating at the frontier	6%	94%	—	17
Delivering independently	—	93%	7%	14

Segment

Too slow

Well paced

Too fast

Applying in practice

95%

—

Learning and exploring

75%

20%

Operating at the frontier

94%

—

Delivering independently

—

93%

Cohort	0	1	2	3	4	5	6	7	8	9	10
Cohort 27			1	1		1	1	9	15	7	21

Cohort

Cohort 27

1	2	3	4	5	Mean
—	5%	20%	43%	32%	4.02

Mean

—

20%

43%

32%

4.02

1	2	3	4	5	Mean
2%	2%	18%	46%	32%	4.05

Mean

18%

46%

32%

4.05

1	2	3	4	5	Mean
2%	5%	20%	46%	27%	3.91

Mean

20%

46%

27%

3.91

NPS Reason

n=49 responses · grouped by NPS segment

Promotersscore 9–10 · 23 responses

1Easy to follow sessions10

2It has been a great experience, Very useful and well prepares with hands on AI10

3Excellent content10

4Super interesting base foundation.10

5Its good to get a lot of tricks and tips10

6Extremely charismatic team, good food, great pace and really interesting exercises.10

7Learnings10

8Absolute eye opener on som many assumptions and preconceived ideas we have about how LLM s operate and are to be used and refined for use cases10

9Smooth, well presented, informative and more importantly very helpful presenters.10

10Great level of depth. Often these courses don’t touch on detail. Candid discussions around trade offs. No correct answers10

11The training touched real world relevant topics10

12hands on experience and practice that studying for an exam alone would not necessarily give you10

13It completely changed how I think about building AI solutions it was practical, hands-on, and covered concepts like context engineering and evaluation that I’d never been exposed to before.10

14Learning by using the Claude’s capabilities10

15I think it’s not only a getting started program. Even if you use AI in daily basics with your clients in production, you’ll learn tips and hints on how to improve your systems; haiku capabilities, warming up cache, etc10

16It's a good foundation to get up to speed quickly. Then people can take their own time to sharpen their skills.10

17Good breath of knowledge, very well run course. Enjoyable9

18Excellent starter9

19Useful content for working with anthropic products, but also on the path to certification9

20The training session was informative, educative and insightful. It was accompanied by practical examples and use cases.9

21Very beneficial in clearing miss conception about the power truck and compliance of AI especially claude9

22Many things that actually I was not sure about implementing solution using Claude code. It was great experience working on different exercises9

23I loved it! The only reason I didn't give it a 10 is that the content, while great and very well organized, can be a bit basic for many of my colleagues at my company. But I would recomend it anyway9

Passivesscore 7–8 · 22 responses

1Great information, very fast paced8

2Optimisation8

3You learn a lot of stuff8

4Good pace of lessons, engaging with the hands on activities using jupyter8

5Great learnings, especially for engineers working on agentic systems/use cases.8

6A good starter and good depth go material. Some of the notebooks were hard to use and not always that intuitive8

7Good foundation8

8I am a newbie in the domain and I feel to confident to start now.8

9I understand different levers to pull to improve solution accuracy , cost or speed, and appropriate ways to build multi agent systems8

10Very hands on and useful to get into working with Claude Code. Depends on if the colleague is working on project with Claude available.8

11Interesting topics that keeps you up to date with AI8

12I think it can be too basic. But it’s the right level for a basecamp I think8

13I think it gives you a great overview of anthropic capabilities and also it gives you useful tools for the real production environments such as the evals and infernce optimization8

14It was a good mix of hands on and information. However time for the hands on sessions was to short to really get into it. I had to rush too much to keep pace.8

15Interesting content, very hands on.7

16Fun course7

17Depends on knowledge level if helpful. Trained architects - not very helpful. Beginner - very helpful. For myself, I would like ultimate deep dive / black belt training7

18day 2 with evals and inference was really good. end of day 1 needs a little tightening7

19Would be a 10 if there where ready to use sandboxes instead of having to work around our corporate configurations :D7

20Most of my direct colleagues are not developers7

21Good content, but wasn’t what I was expecting. Would like to have built something from scratch - still no idea where to start with that! Would also been better to have a getting started project I could have downloaded and run up in advance, so I wasn’t spending the firm morning talking to my helpdesk to get going7

22This course will help to start building the knowledge and practice around using Claude and its tools like code7

Detractorsscore 0–6 · 4 responses

1Lack of claude cowork, claude design and claude desktop app. Which is what pwc will support antrhopic in selling6

2Depending on colleague5

3Not technical enough for my team3

4Have been working with the stuff for a while, was hoping for a deeper dive. This is a personal sentiment though,2

Most Valuable

What one takeaway will you share with a colleague or client? · n=37 responses

1Model is not the problem check your prompts

2Choosing the right model and prompt

3Less is sometimes more when it comes to choosing the model.

4The ability to carry out jobs hands off

5Inference process and optimisation

6Evals are the most important thing. Many people dont know about it much

7The relative costs of models.

8Managed agents

9Read more

10Context engineering

11The speed of build and how to improve the accuracy keeping costs low

12Hackathon demo

13ROI - Generate Value with cost effectiveness at the same time

14Eval

15Excellent stuff. Explore ….

16There absolutely are low hanging fruits in saving tokens

17If something is wrong it is not usually the model, it's the stuff around it

18It’s not always about improving model , it’s about improving the implementation of the model

19All models are good what you do with them makes the difference

20It’s not always the model that’s at fault. 🙂

21evals are required before taking AI apps to production

22Interview yourself and build a skill Context management is everything and it’s all tradeoffs The prompt rescue notebook. Best tutorial notebook structure and style I’ve seen. Very intuitive.

23Do not just switch to the advanced model and hope everything will workout better

24A good framework can make Haiku give Sonnet results.

25context engineering as a concept

26AI success depends more on the quality of the context you provide than on the model you choose.

27That system prompts can be repaired if you look at the entire AI solution.

28Claude can do amazing things

29AI can make your life easier

30AI prompting

31More powerful models != better model choice

32The solution is rarely changing the model, but the harness around it

33Half is the model, half is the context/harness. Sometimes, mostly maybe, we have to fix that instead of change model

34The same max tokens tip

35even if you hit 100% on evals you may be using a model too expensive and inference optimization is the other useful tool to pair with evals before production

36Basically the same things I said in the previous question

37I'll do a prsentation about the basecamp and continue to distribute knowledge how to use agenti ai efficiently

30-Day Intentions

What will you build for yourself or apply at work? · n=40 responses

1Personal productivity agent

2Swarm workflow

3Better evaluation !!

4All really

5Try out dynamic workflows for my use cases.

6Add api calls into workflow instead of purely using as a development aid

7Certify under architect.

8Managed agents, inference optimization

9User training agent

10I will carry on with the hackathon app

11Fix the prompts built so far

12EvalOps repo plugin

13Personal Finance App with Claude Code

14Still confused.

15Eval

16Read through and exprement

17For sure exploring the advantages of the claude api vs bedrock or others, like the caching

18I will be organising my data with cowork and building agent swarms to assist with day to day management duties. I will then be designing solutions for clients and pursuing foundation certification

19A demo

20Complete the Hackathon to have a UI

21An AI compliance system

22Use specific skills to tune output. build evals to verify quality

23knowledge base

24a persistent + growing memory agent

25Evals, cost and latency comparisons

26The advisor concept for a smaller model.

27use of sub-agents with better curated, trimmed and targeted context

28Apply context engineering and evaluation techniques to improve our AI agents and deliver more reliable outputs.

29I want to leverage MCPs and agents to do life admin tasks (meal prep ideas or even doing a wardrobe audit)

30Using Claude everyday

31I don't really work with AI

32Incorporating evals in work flow taking away some brilliant ideas

33Some multi agent personal workflows

34Cache warm up, and give it another opportunity at little haiku as I think now that it’s more capable than we thought before

35Just keep doing as now, building tools and skills that will help me and my team to improve efficency in coding

36The caching with max tokens set to zero

37evals and inference optimization

38All three hecathon demos and certification

39Mainly the things I learned that are tricky to learn by yourself or they are counter intuitive. These two: - The prewarming of the cache before launching parallel requests. - Using Haiku confidently with the right harness. It can be almost as capable as bigger models

40Agents whon can memorize and improver their skills

Day 1 responses	72
Day 2 responses	56 (78% of Day 1)
Matched pairs	53 (95% of Day 2)
Orgs resolved	72 of 72 respondents matched to named org
Day 2-only respondents	3 (5.4%) — suppresses NPS by 1 points; no Day 1 data, excluded from persona & confidence analysis

London