← Home
Partner Basecamp · Cohort 29

San Francisco

August 24–25, 2026
+48
Net Promoter Score
97
Day 1 Responses
96%
Matched Pairs
87%
Day 2 Response Rate
Programme Satisfaction
Positive NPS of +48 — promoters outweigh detractors with moderate passive presence.
+48NPS · n=84
58% Promoters31% Passives11% Detractors
95% CI: +33 → +62  ·  True NPS lies within this range with 95% confidence (n=84 respondents)
Confidence Arc
Moderate end-of-Day-1 build confidence — cohort is progressing; Day 2 has room to close remaining gaps (mean 3.8/5).
3.8end-of-D1 build / 5
4.0D2 Design / 5
4.1D2 Commercial / 5
4.2D2 Build / 5
Audience
97 participants across 16 organisations — Transformation Lead majority with applying in practice the most common AI experience level.
Top Organisations
DXC Technology (19) PwC (14) Deloitte (14) Accenture (13) Cognizant (6)
Voice of Participant
What participants said — dominant theme from each NPS segment.
Promoters
Hands-on exercises with expert speakers delivered innovative, practical AI concepts participants could immediately apply to client work.
Passives
Valuable content suited experienced users, but mixed skill levels made foundational sections feel redundant for those with prior Claude experience.
Detractors
Across a small number of responses (n=8), the recurring ask was for more structured teaching and explicit key takeaways, particularly for non-technical participants.
Segment Finding
Ascendion (NPS +100, n=5) and PwC (NPS -71, n=7) bracket the cohort. Ascendion was carried by the depth of the agentic-development content and Claude's tool-integration story for client POCs — both verbatim 10s came from Architects at opposite ends of the experience range, one 'delivering independently' and one still 'learning and exploring.' A third Ascendion respondent scored 9, calling the programme 'a great start for a consultant in agentic solutions' but not next-level for someone already building — a visible ceiling effect even inside the top-scoring org. PwC, at the other end, had 5 detractors of 7, and four of those five verbatims name the same cause: not enough structured instruction, upfront exercise framing, or explicit key takeaways for participants who aren't deeply technical. One detractor separately cites unresolved firm IT issues that blocked participation; one is a frontier-level Developer for whom the material was simply too familiar. Both PwC passives raise the same framing gap, so the signal is content-fit and scaffolding rather than content quality — the same gap that drove C28's NPS drag, now visible in a second cohort. Both segments are small (n=5, n=7); directional only.
Analyst Note
Promoter open text (n=37) clusters on five themes, by frequency of mention: learning gain and new capability (13), hands-on exercise design (10), breadth of Claude and agentic coverage (8), instructor quality and engagement (5), and applicability to client work (5). Representative responses: 'Immersive and breadth covered on agentic based development and diagnosis.' 'It was interactive and the instructors were very helpful.' 'The basecamp didn't feel boring for a single sec. The way the program was designed was so interactive and hands-on, making it like a game we are playing.' From a self-described beginner: 'I was nervous coming in as a beginner. But now I feel much more confident in my abilities. Learned a lot of techniques that I will have to expand on when I get back.'
Survey & Data Quality
Survey & Data Quality
Day 1 responses97
Day 2 responses84 (87% of Day 1)
Matched pairs81 (96% of Day 2)
Orgs resolved97 of 97 respondents matched to named org
Day 2-only respondents3 (3.6%) — no NPS data; no Day 1 data, excluded from persona & confidence analysis
Organisations
All Day 1 respondents · n=97
DXC Technology19 (20%)
PwC14 (14%)
Deloitte14 (14%)
Accenture13 (13%)
Cognizant6 (6%)
Aimpoint Digital5 (5%)
Ascendion5 (5%)
NEC5 (5%)
AWS4 (4%)
IBM3 (3%)
Forgd.AI2 (2%)
EPAM2 (2%)
Hellman & Friedman2 (2%)
Bain1 (1%)
McKinsey1 (1%)
Wipro1 (1%)
Function × Seniority
All Day 1 respondents · n=97
FunctionPractitionerSr PractitionerMgr / Sr MgrDirector+Partner / MDTotal
Engineering101093133
Architecture6661–19
Business Leadership1158520
Project / Engmt13461125
Experience Profile
AI experience level · n=97 · % of row
By Persona
SegmentExploringApplyingDeliveringFrontiern
Transformation Lead22%56%18%4%45
Developer18%45%12%24%33
Architect11%53%21%16%19
By Seniority
SegmentExploringApplyingDeliveringFrontiern
Practitioner (0-4 years in role)23%67%3%7%30
Manager or Senior Manager12%62%15%12%26
Senior practitioner (5–9 years)19%38%24%19%21
Director, Senior Director, or Principal15%31%38%15%13
Partner, Managing Director, or Executive29%29%14%29%7
How would you describe your current experience with AI tools prior to Basecamp?
Learning and exploring
18 (19%)
Applying in practice
50 (52%)
Delivering independently
16 (16%)
Operating at the frontier
13 (13%)
Prior to today, how much had you worked with Claude or the Anthropic API?
Not at all
8 (8%)
A little
42 (43%)
Regularly
47 (48%)
Did the depth land for this audience?
Technical depth perception · n=97 respondents · % of row
By Persona
SegmentToo basicAbout rightToo advancedn
Transformation Lead11%67%22%45
Developer15%85%—33
Architect21%74%5%19
By AI Experience Level
SegmentToo basicAbout rightToo advancedn
Applying in practice16%70%14%50
Learning and exploring6%78%17%18
Delivering independently6%88%6%16
Operating at the frontier31%69%—13
Overall depth distribution
Too basic
14 (14%)
About right
72 (74%)
Too advanced
11 (11%)
Did the pace work across the room?
Session pace perception · n=97 respondents · % of row
By Persona
SegmentToo slowWell pacedToo fastn
Transformation Lead9%69%22%45
Developer—85%15%33
Architect11%68%21%19
By AI Experience Level
SegmentToo slowWell pacedToo fastn
Applying in practice6%70%24%50
Learning and exploring—89%11%18
Delivering independently12%69%19%16
Operating at the frontier8%77%15%13
Overall pace distribution
Too slow — could have covered more
6 (6%)
Well paced
72 (74%)
Too fast — not enough time to apply
19 (20%)
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=97 · 1–5
Mean end-of-D1 build confidence by Persona
Developer4.15/5 · n=33
Architect4.11/5 · n=19
Transformation Lead3.56/5 · n=45
Programme mean: 3.9/5 · 20 cohorts
Mean end-of-D1 build confidence by AI Experience Level
Operating at the frontier4.77/5 · n=13
Delivering independently4.12/5 · n=16
Applying in practice3.64/5 · n=50
Learning and exploring3.61/5 · n=18
Overall end-of-D1 build confidence distribution
1
4 (4%)
2
8 (8%)
3
18 (19%)
4
34 (35%)
5
33 (34%)
How relevant was today's content to your current role?
Content relevance rating · n=97 respondents · 1 = not relevant, 5 = highly relevant
Mean relevance by Persona
Architect4.47/5 · n=19
Developer4.18/5 · n=33
Transformation Lead3.36/5 · n=45
Mean relevance by AI Experience Level
Operating at the frontier4.38/5 · n=13
Delivering independently3.88/5 · n=16
Learning and exploring3.78/5 · n=18
Applying in practice3.74/5 · n=50
Overall relevance distribution
1
3 (3%)
2
6 (6%)
3
27 (28%)
4
27 (28%)
5
34 (35%)
How likely are you to recommend attending this programme to a colleague?
n=84 Day 2 respondents · 87% of Day 1
Score distribution · 0 = not at all likely · 10 = extremely likely
Cohort012345678910
Cohort 291121411151831
■ Detractors 0–6■ Passives 7–8■ Promoters 9–10
NPS Breakdown
58%
31%
11%
NPS +48 (n=84)
Promoters (9–10)Passives (7–8)Detractors (0–6)
Programme mean: +42 · 20 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)
Ascendion
100%
0%
0%
NPS +100 (n=5)
Accenture
85%
15%
0%
NPS +85 (n=13)
Aimpoint Digital
75%
25%
0%
NPS +75 (n=4)
DXC Technology
71%
29%
0%
NPS +71 (n=17)
Cognizant
67%
33%
0%
NPS +67 (n=6)
Deloitte
58%
33%
8%
NPS +50 (n=12)
NEC
40%
60%
0%
NPS +40 (n=5)
PwC
0%
29%
71%
NPS -71 (n=7)
By Persona
Promoters (9–10)Passives (7–8)Detractors (0–6)
Architect
65%
35%
0%
NPS +65 (n=17)
Developer
61%
29%
10%
NPS +52 (n=31)
Transformation Lead
54%
27%
18%
NPS +36 (n=33)
Programme means · 20 cohorts: Architect +50 · Developer +48 · Transformation Lead +36
By AI Experience Level
Promoters (9–10)Passives (7–8)Detractors (0–6)
Learning and exploring
71%
24%
6%
NPS +65 (n=17)
Applying in practice
67%
21%
12%
NPS +55 (n=42)
Operating at the frontier
46%
36%
18%
NPS +27 (n=11)
Delivering independently
27%
64%
9%
NPS +18 (n=11)
Programme means · 20 cohorts: Applying in practice +45 · Learning and exploring +44 · Delivering independently +43 · Operating at the frontier +39
What is the main reason for your score?
n=66 responses · organised by NPS segment
Promoters (score 9–10)· 37 responses
Hands-on exercises with expert speakers delivered innovative, practical AI concepts participants could immediately apply to client work.
10“As automation engineer , we have to think lot of tools while doing POCs for new projects.After working with claude we can definitely integrate other tools with claude and it will give real time analysis.”
Passives (score 7–8)· 21 responses
Valuable content suited experienced users, but mixed skill levels made foundational sections feel redundant for those with prior Claude experience.
7“1) I feel like a lot of this could have been completed as an online exercise, and I’m a bit disappointed there haven’t been more structured lessons or clear key takeaways tied to each exercise.   2) It felt like we were sent into some of the exercises pretty quickly without enough upfront context, examples, or key terms to help guide the solution. For example, with 01_evals, it would have been helpful to give a little more direction on how to approach prompting with the Claude Code plugin in VS Code. Something like: “Understand the agent, set up the eval harness, and take three passes to improve the robustness of both the eval harness and the agent.” Or even more simply: “Make the agent as robust as possible and design a world-class eval, then explain what you changed and why.”   The skill level in the room is also pretty broad, so I think accommodating that range would be valuable feedback for future Basecamp sessions. I also think each exercise should do a better job of reinforcing the key takeaway and why the exercise matters, so people leave understanding the underlying concept rather than just completing the task.”
Detractors (score 0–6)· 8 responses
In-person format underutilized for self-directed activities; non-technical participants needed more facilitation and prompt-engineering guidance beyond conceptual lectures.
4“Facilitators were helpful 1-1 and brought good energy. The sessions, except hackathon, felt like something I could do online and in some ways operated on their own without teaching (e.g. Claude literally responded saying you won’t learn this way). As a nontechnical person, would have appreciated seeing prompts in GitHub vs code. Teach me the types of things I can and should ask such that come hackathon, I am inspired and can do on own.”
End-of-Programme Confidence
n=84 respondents · 1 = not confident, 5 = very confident
How confident are you in your ability to design an evaluation for an AI solution?
12345Mean
2%5%14%43%36%4.05
Programme mean: 4.1/5 · 20 cohorts
Leaving today, how comfortable are you advising clients on AI when questions arise?
12345Mean
1%2%15%44%37%4.13
Programme mean: 4.2/5 · 20 cohorts
How confident are you in your ability to lead a conversation about Anthropic and Claude with a client?
12345Mean
1%5%12%40%42%4.17
Programme mean: 4.2/5 · 20 cohorts
Confidence Trajectory by Persona
D1→D2 delta · apply-AI confidence · matched pairs only · n=81
Positive delta = confidence grew · negative = dropped
Architect (n=17)
+0.53
Developer (n=31)
+0.16
Transformation Lead (n=33)
+0.39
What one takeaway will you share with a colleague or client?
n=50 responses
Participants repeatedly emphasized evaluation frameworks as the cornerstone of responsible AI implementation in client projects.
“Eval, eval, eval”
Consultants recognized that model selection matters less than thoughtful implementation strategy and comparative effectiveness testing.
“It's not the model it's how we implement it”
Participants valued grounding agent development in software engineering fundamentals and multi-agent architectural patterns.
“Start with engineering first principles”
Consultants shifted focus from faster coding to designing AI solutions that measurably improve entire business processes and outcomes.
“AI can create the most value when we focus not just on coding faster, but on improving entire business processes and outcomes.”
Participants applied practical frameworks like client checklists to translate inference-optimization concepts into deliverable client work.
“The key points from our inference-optimization exercise with the client checklist”
What will you build for yourself or apply at work in the next 30 days?
n=56 responses
Participants want to build practical agent systems, from stock analysis to hiring workflows, applying agentic patterns to their SDLC.
“Agentic SDLC workflows”
Participants need to implement evaluation systems and deterministic gates to validate agent outputs and optimize performance reliably.
“Creating gates and deterministic evals”
Participants plan to identify better ways to leverage existing AI applications and present validated use cases to their clients.
“Presenting good use case to clients”
Participants intend to return to clients armed with Claude capabilities and governance models to integrate AI into active projects.
“I am going to go back to my client and present them with what Claude can do and where it fits in their project.”
Participants want to revisit labs and training exercises to deepen understanding and internalize the levers available for AI implementation.
“Go over labs again, to get a deeper understanding and get some reps in to better memorize the various levers we can pull.”
Partner Basecamp Survey Report · Cohort 29 · Generated by analyze.py v2.5
NPS Reason
n=66 responses · grouped by NPS segment
Promotersscore 9–10 · 37 responses
1Very informative10
2Great speakers, fun 2 days10
3Vibes.10
4Very innovative with hands on exercises10
5Immersive and breadth covered on agentic based development and diagnosis10
6To understand Claude AI breadth useful for my role10
7New concepts, insights from anthropic team10
8Detailed information.. good presentation10
9It was interactive and the instructors were very helpful10
10The handson session on Agentic AI architecture/framework and the usage of Claude to build real prod ready env. Talk to real Anthropic teams about their experiences and challenges.10
11I was nervous coming in as a beginner. But now I feel much more confident in my abilities. Learned a lot of techniques that I will have to expand on when I get back.10
12Great content, atmosphere, and instruction. Very engaging imo10
13It's informative in short time you get to have more overview about Claude code10
14The basecamp didn’t felt boring for a single sec. The way program was designed was so interactive and hands on experience making it like a game we are playing on.10
15very informative learning for someone not already familiar with claude code.10
16The training provided me with knowledge on Claude some which I was not aware off and there was a lot of learning from this session.10
17The content was good and met good crowd10
18Hands on. Tailored to what we’re seeing in the marketplace. I found that I have a whole bunch of aha moments, good examples, and mini learnings to apply.10
19It's important that my peers have the level of understanding of Claude's capabilities and the level of proficiency at agent creation and modification gained here today.10
20As automation engineer , we have to think lot of tools while doing POCs for new projects.After working with claude we can definitely integrate other tools with claude and it will give real time analysis.10
21Really enjoyed hands on learning, Clarissa super engaging as instructor for the last two days. We will be sending more engineers in the future.10
22Learned a lot, made friends, good networking opportunity10
23I learned a lot from the exercises.10
24Good content9
25Relevant content9
26Got very useful insights and interesting information9
27This was organized very well with lots of handson and insightful discussion9
28I find it intersting and great learning experience.9
29Was rigorous in technical training9
30Good food, good company and knowledge sharing.9
31Great deep dive into Claude. Would recommend giving learners more context on how they should prepare with some Linux knowledge and API key setup9
32A lot of skills packed into just two days. This will be a great source of inspiration to bring back to my clients and keep building business value.9
33This is a great start for a consultant in agentic solutions. But cannot say 10/10 cuz as its not next level for someone already doing this stuff9
34Regardless of the level of AI familiarity and skill, the basecamp prepares the novice to the advanced on core concepts9
35Good mix of salesy material, technical knowhow, hands-on labs, and interactive q&a. Big shout out to Carissa for remembering our names!9
36Great blend of informational slides, hands on activities in a fun and low pressure environment created for learning with practical use cases9
37High9
Passivesscore 7–8 · 21 responses
1Good contents8
2残念ながら全ては理解できなかった。でもチームのメンバーとコミュニケーションとりながら楽しく進められたから。 あと、クロード先生は日本語がわかるからワラ8
3Very deep and handson8
4Very useful program, especially if you have some initial claude code background8
5Great to spend time playing around with Anthropic tools8
6I learned a lot8
7It was nice to work in the system and get a better understand of what tools are at my disposal8
8The score have been 10 if the assignment/lab were connected to each other and part of a same application/concept.8
9good mix of technical work and how to bring it back to clients.8
10Good hands on experience but some elements of the workshop required a deeper understanding of python code to review and the time for topics on day 2 for those was not sufficient8
11This program is good for developer wants enforce Agent-Driven Development.8
12Learning practical methods for evaluating agents and optimizing their performance was the biggest takeaway for me8
13Second day was more informative.8
14Very informative and interactive7
15Covers relevant use cases that can be used in a client setting. Would like it to cover more technical detail.7
16Day 2 was better. I wish there was more focus on the details of why we do things and best practices. The tasks we did were easy enough to just give Claude access to the file and say “fix it”7
17Solid content coverage, but mixed skill level audience leaves day 1 feeling a tad remedial for more experienced users—score would be higher for someone who really needs a 101 vs a 2017
18I think it’ll vary by colleague. Will be super useful to some. But for the deeply technical folks, some of the content might be straightforward.7
19Nice hands-on exercises and even better vibes. Would have liked a bit more instruction and POV on the latest thinking / recommendations rather than just turning people loose on the Jupyter notebook exercises.7
20Well covered topics, but not enough time for activities7
211) I feel like a lot of this could have been completed as an online exercise, and I’m a bit disappointed there haven’t been more structured lessons or clear key takeaways tied to each exercise.   2) It felt like we were sent into some of the exercises pretty quickly without enough upfront context, examples, or key terms to help guide the solution. For example, with 01_evals, it would have been helpful to give a little more direction on how to approach prompting with the Claude Code plugin in VS Code. Something like: “Understand the agent, set up the eval harness, and take three passes to improve the robustness of both the eval harness and the agent.” Or even more simply: “Make the agent as robust as possible and design a world-class eval, then explain what you changed and why.”   The skill level in the room is also pretty broad, so I think accommodating that range would be valuable feedback for future Basecamp sessions. I also think each exercise should do a better job of reinforcing the key takeaway and why the exercise matters, so people leave understanding the underlying concept rather than just completing the task.7
Detractorsscore 0–6 · 8 responses
1Lots of tech issues not yet solved for our firm. Also felt like there wasn’t a huge emphasis on key takeaways from each exercise - I had to prompt Claude to summarize what I was doing and why5
2As someone who has been building with Claude and is certified, a lot of the material that was covered had concepts I was familiar with. This would be a very good program for professionals who are newer to building in the Gen AI space.5
3Less technical than most of my colleagues would benefit from. I did learn some things and it was helpful to walk through some of the exercises but have experience with most everything discussed via my casework.5
4The lecture was helpful to learn concepts, but the activities not so much.5
5Facilitators were helpful 1-1 and brought good energy. The sessions, except hackathon, felt like something I could do online and in some ways operated on their own without teaching (e.g. Claude literally responded saying you won’t learn this way). As a nontechnical person, would have appreciated seeing prompts in GitHub vs code. Teach me the types of things I can and should ask such that come hackathon, I am inspired and can do on own.4
6Felt that it could have been more interactive with instructors. All activities were done on own which could have been done from home. Being in person could have been taken advantage of more3
7Didn’t feel organized. Trainers were not as technical as I expected. It was more self learning which I would not have had to travel for. Too many people in training to ask real questions and get real learning. Would have loved more breakout sessions. Supporting theory. Snacks. I attended OpenAI boot camp last week. Much better.3
8Not enough lessons, lecturing, concepts or key takeaways for folks who arent as technical1
Most Valuable
What one takeaway will you share with a colleague or client? · n=50 responses
1Eval, eval, eval
2Agent review
3Eval system
4Start with engineering first principles
5Multi agent framework
6Evaluations!
7It’s not the model it’s how we implement it
8Evals, approach to diagnosis issues with agents in prod
9How powerful Evals are.
10Better accuracy does not necessarily mean increased costs.
11Changing Model is not always the solution
12Evals are extremely powerful.
13Exploring Claude and comparing all models and their effectiveness
14Hopefully the ability to get started with Claude
15How important evals are.
16Looping and LLM as judge.
17I’ll share details regarding Claude tags
18Think bigger.
19Claude is secured , fast, efficient.
20Evals
21How the agents are built-in
22The importance of evals and managing cost, model escalation, etc.
23You don’t always have to spend more money to get more value
24Evals
25Spend time building good eval for the AI solution.
26a lot of optimization can be done on prompts, context etc that can really bring out best value from the agents.
27Structured evals
28claude code can save a lot of time
29the eval framework
30AI can create the most value when we focus not just on coding faster, but on improving entire business processes and outcomes.
31Very good start. Nice excercises
32Can help assess drift with evals tagged to release
33Learn Claude it can do almost everything
34model selection for different tasks to optimize spend
35Don't be afraid to build using claude, make sure you have good evals
36LLM as a judge is a powerful eval when used correctly.
37It’s not always the model.
38Think beyond the model
39How fast token usage increases with agents.
40Evals evals evals
41The better understanding I got of the evaluation and remediation process.
42Options are open, but stay focused on how to choose and apply the levers and decision making appropriately. Test first, apply evals and then be thorough to review before going live...
43Clarity,Reasoning,Insight,Precision,Orchastration
44The key points from our inference-optimization exercise with the client checklist
45Evals and inference optimization
46Evals! Numbers speak louder than words
47One agent doing five jobs in sequence looks like a system but behaves like one long answer. Splitting the work so that one agent produces and another challenges it is what makes the output trustworthy
48Understanding models based on behavior types is key to starting initial convos with clients.
49Good evaluation isn’t optional — it’s the foundation for building AI agents that are both reliable and cost-effective.
50Importance of evaluating
30-Day Intentions
What will you build for yourself or apply at work? · n=56 responses
1I’ll keep building ideas
2Inference approaches
3Eval system
4Eval and optimization
5Creating gates and deterministic evals
6A stock market analysis agent
7Pokemon Announcer
8Try to leverage existing AI applications in a better way
9Evals
10Hiring agent from Accenture
11My own harness
12Anything that I can think of! I am ready to design and build solutions based on what I learned from this program.
13Agentic SDLC workflows
14Multi-agent frameworks
15Evals
16Presenting good use case to clients
17For myself - flight tracker. For work - plan to reduce repeat activities to increase productivity
18Evil framework
19Just get started and try!
20My goal is to become More efficient when building client demos with Claude.
21Go over labs again, to get a deeper understanding and get some reps in to better memorize the various levers we can pull.
22Agentic SDLC that was presented by Swarm and Order team during the hackathon
23Token management for AI systems that we are building. Personal interest apps on the side.
24Planning to live application.
25Skills and evals
26How to integrate with SAP and use data from sap build more agents
27Agent dev loop system for task delegation to execute plans for new build and bug resolution.
28An agent to help with my Smartsheets updates and an onboarding agent t
29Autonomous agents for my clients solving various problems we already know of
30A good eval framework.
31Decision Time Machine
32A multiagent voice system
33prototype for tech stack upgrades
34rollout governance models, agents!!!
35I’ll apply what I learned to build a small AI-powered SDLC workflow that improves team productivity and helps identify opportunities to deliver more business value for my clients.
36Im planning to redesign a dashboard at work using multiagent framework
37Hoping to build out evals
38I am going to go back to my client and present them with what Claude can do and where it fits in their project.
39Evals
40Infra sec ops governance pipeline
41A personal project for to manage repairs to my car
42It’s all the micro learnings to apply to existing, inflight projects and prototypes.
43We have multiple client POCs we can start building
44Agent for applied Ai workshops.
45Maybe a spotify playlist rules engine
46I expect to build an agent or two to fill gaps in our current offering space.
47Dig deeper into levers, evaluate and cost optimization for the context. Help my clients choose appropriate models for their context and workload.
48We have our own AI platform we will try to configure it with claude. We will try to automate applications using it.
49Leverage the knowledge from our Day 2 to build an eval
50Claude code evals will be critical for the our client use cases from more advanced AI agents and AI workflows. Looking forward to building with these skills
51Too many ideas to decide lol
52H
53apply agent fixing cycle to my agents.
54I plan to re-run through training exercises from day 1 and day 2.
55I want to build an evaluation framework and start using it across my team and organization. I’ll also focus on building agents with cost and performance in mind.
56Context engineering