AI Manifesto

How Crema works in an AI-accelerated world

We use AI to amplify creativity, accelerate delivery, and expand what's possible—but we own the outcomes. Every insight validated. Every decision human-led. Every client trust protected. Because moving fast only matters if you're moving in the right direction.

Our perspective

This is a pivotal moment. AI represents a cultural and technological inflection point comparable to the internet—democratizing capabilities that were once specialized. As information becomes universally accessible, the differentiator shifts from what you know to how you apply it.

Crema will leverage what's necessary to deliver results. We're pragmatic, not reactive. Just as we've championed emerging technologies throughout our history, we'll leverage AI to deliver results—not for its own sake. When everyone has access to the same models, expertise and judgment become the advantage. AI should amplify our contributions, not replace them.

We're facing this inflection point thoughtfully. Pace matters, but so does direction. Moving too fast risks instability; too slow means missed opportunities. Success requires balancing speed with wisdom, understanding tradeoffs, and maintaining what makes work exceptional: the ability to navigate complexity and shape it into something brilliant.

We're committed to our top-quality standards.
Crema remains a premium, relational, strategic partner. Our cross-functional expertise, deep client relationships, and commitment to delivering innovative solutions don't change—they're what help clients navigate this transformation responsibly and successfully.

AI is here.

We can’t really go back from this moment. While some see an exciting frontier, we understand not everyone feels ready. At Crema, we’re leaning in because we see this as an inflection point. One that will amplify our potential in ways we're only just beginning to understand.

We’re committed to bringing you along this learning journey, optimistic about how it will improve our capacity and creativity so that we can do the best work of our lives.

This is a peek behind the curtain on how our craft teams are experimenting and ultimately implementing a variety of colorful use cases for AI.

SPARK

Deciding where to experiment.

Instead of throwing AI at everything to see what sticks, we built a framework for it. If AI could help with any of these problems, then we proceed with an experiment.

S
cale
High volume or time intensive task
P
attern
Repeatable structure or behavior
A
mbiguity
Needs perspective or ideation
R
edundancy
Been done before, often
K
nots
Blocks or slows down people
Spark

Deciding where to experiment.

Instead of throwing AI at everything to see what sticks, we built a framework for it.

Scale
High volume or time intensive task
Pattern
Repeatable structure or behavior
Ambiguity
Needs perspective or ideation
Redundancy
Been done before, often
Knots
Blocks or slows down people

If it checks all of the boxes, then we proceed with an experiment.

Experiments

Crema AI experiments

At Crema, an explicit expectation has been set that every team member should be experimenting with and validating AI-assisted workflows.

We set a clear timeline for all craft teams to share learnings with the broader organization. Here are just a few examples of what was discovered.
Project
Craft team
Tools used
Getting a complex app running locally without the headache
  • Goal: Set up a full application stack (SQL Server, .NET API, Azure functions, React front end) to run locally on any developer's machine, eliminating dependency on slow virtual desktops and shared dev environments.
  • Result: After spending most of a day working with Claude in Cursor, pointing it to config files and explaining how pieces connected, we got a fully containerized Docker setup that anyone can spin up. New developers onboard faster, everyone has their own test data, and teams can now make API and front end changes simultaneously without waiting on deployment pipelines. The time savings have been massive.
Craft team:
Development
Tools used:
Claude, Cursor
Development
Claude
Cursor
Building test cases for a natural language search feature
  • Goal: Create a comprehensive test suite for a natural language search engine that needed to interpret real world user queries like "show me parcels over 200 acres in Austin, TX" or "I'm looking for big land near Austin."
  • Result: Generated 150 test prompts covering basic queries, natural variations, edge cases, misspellings, and contradictory logic. Each prompt was tested by three different people. This organized approach helped us run a four hour team testing session that caught critical issues before client release. We plan to keep using this for features that rely on natural language interpretation.
Craft team:
Testing
Tools used:
Claude,
Testing
Claude
Understanding team dynamics across organization levels
  • Goal: Synthesize interview data from a product team assessment to identify how perspectives align or differ between organizational levels.
  • Result: Used ChatGPT to organize interview notes, then prompted it to build a comparison matrix across org levels. This revealed gaps and patterns we might have missed or would have taken much longer to find manually. Being able to see the data through this comparison lens was new and pointed us in a better direction as we prepared findings for the client.
Craft team:
Strategy
Tools used:
Claude
Strategy
Claude
Reverse engineering and rebuilding an app from scratch
  • Goal: Reverse engineer a messy, vibe coded project to understand its structure, then test whether a clean rebuild using better AI prompting could produce production quality code.
  • Result: Used Cursor to map the original app into stories and epics, then started fresh with a data first approach. Basic development tasks were roughly 70% faster, especially for components, tests, and test data. When AI got something wrong, correcting it and having it generate documentation helped future tasks. Refactoring UI elements was much quicker than doing it manually. This is definitely a practice worth continuing, particularly for developer onboarding and understanding unfamiliar codebases.
Craft team:
Development
Tools used:
Cursor, ChatGPT
Development
Cursor
ChatGPT
Automating meeting registration and follow up workflows
  • Goal: Learn to use Zapier to automate the repetitive, time intensive process of managing event registrations, which involved monitoring submissions, evaluating applicants, sending notifications, and updating multiple systems.
  • Result: Built a workflow that handles the entire registration pipeline: notifying the Slack channel of new Typeform submissions, analyzing job functions to determine acceptance, sending status emails, adding accepted participants to Google calendar invites, and updating the Notion database. This eliminated constant browser switching and the risk of things falling through the cracks, freeing up time that previously went to tasks requiring little cognitive function but plenty of attention.
Craft team:
Operations
Tools used:
Zapier
Operations
Zapier
Experiments

Crema AI Experiments

At Crema, an explicit expectation has been set that every team member should be experimenting with and validating AI-assisted workflows.

We set a clear timeline for all craft teams to share learnings with the broader organization. Here are just a few examples of what was discovered.
Project
Craft team
Tools used
Getting a complex app running locally without the headache
  • Goal: Set up a full application stack (SQL Server, .NET API, Azure functions, React front end) to run locally on any developer's machine, eliminating dependency on slow virtual desktops and shared dev environments.
  • Result: After spending most of a day working with Claude in Cursor, pointing it to config files and explaining how pieces connected, we got a fully containerized Docker setup that anyone can spin up. New developers onboard faster, everyone has their own test data, and teams can now make API and front end changes simultaneously without waiting on deployment pipelines. The time savings have been massive.
Craft team:
Development
Tools used:
Claude, Cursor
Development
Claude, Cursor
Building test cases for a natural language search feature
  • Goal: Create a comprehensive test suite for a natural language search engine that needed to interpret real world user queries like "show me parcels over 200 acres in Austin, TX" or "I'm looking for big land near Austin."
  • Result: Generated 150 test prompts covering basic queries, natural variations, edge cases, misspellings, and contradictory logic. Each prompt was tested by three different people. This organized approach helped us run a four hour team testing session that caught critical issues before client release. We plan to keep using this for features that rely on natural language interpretation.
Craft team:
Testing
Tools used:
Claude,
Testing
Claude
Understanding team dynamics across organization levels
  • Goal: Synthesize interview data from a product team assessment to identify how perspectives align or differ between organizational levels.
  • Result: Used ChatGPT to organize interview notes, then prompted it to build a comparison matrix across org levels. This revealed gaps and patterns we might have missed or would have taken much longer to find manually. Being able to see the data through this comparison lens was new and pointed us in a better direction as we prepared findings for the client.
Craft team:
Strategy
Tools used:
Claude
Strategy
Claude
Reverse engineering and rebuilding an app from scratch
  • Goal: Reverse engineer a messy, vibe coded project to understand its structure, then test whether a clean rebuild using better AI prompting could produce production quality code.
  • Result: Used Cursor to map the original app into stories and epics, then started fresh with a data first approach. Basic development tasks were roughly 70% faster, especially for components, tests, and test data. When AI got something wrong, correcting it and having it generate documentation helped future tasks. Refactoring UI elements was much quicker than doing it manually. This is definitely a practice worth continuing, particularly for developer onboarding and understanding unfamiliar codebases.
Craft team:
Development
Tools used:
Cursor, ChatGPT
Development
Cursor, ChatGPT
Automating meeting registration and follow up workflows
  • Goal: Learn to use Zapier to automate the repetitive, time intensive process of managing event registrations, which involved monitoring submissions, evaluating applicants, sending notifications, and updating multiple systems.
  • Result: Built a workflow that handles the entire registration pipeline: notifying the Slack channel of new Typeform submissions, analyzing job functions to determine acceptance, sending status emails, adding accepted participants to Google calendar invites, and updating the Notion database. This eliminated constant browser switching and the risk of things falling through the cracks, freeing up time that previously went to tasks requiring little cognitive function but plenty of attention.
Craft team:
Operations
Tools used:
Zapier
Operations
Zapier
Amplify human capacity
  • Enable people to do more good work better more quickly
  • Apply AI to improve contributions across many dimensions of humanity: creativity, productivity, perspective, accessibility, and more.
  • Significant change like this requires new skills, perspectives, and investments from the business down to its employees. As we employ new technologies, we continue our commitment to training and empowering those who use them.
  • Humans can only act from the context they have and AI can dramatically expand perspectives and understanding quickly.
Aim for impact
  • Shorten the time it takes to shape, deliver, and learn
  • Don’t simply use AI for the sake of itself
  • Iterate more quickly
  • Experiment often and find a way to bring it back to the work
  • Prioritize solutions that deliver tangible business outcomes
Own the outcomes
  • Trust and validate everything: AI-enabled tools are not responsible for the outcomes provided to our clients — Crema is. Therefore it’s up to us to verify the legitimacy and overall quality of information received via AI tools.
  • AI should not dictate our direction, but help us pursue it more clearly and quickly
  • Ensure AI implementation enhances rather than replaces human judgment and expertise
  • Just as AI can expand our perspectives, it can greatly extend our skillsets, but we still bear the burden of quality
  • We will be clear and forthright about where & how we’re using AI
Protect what matters
  • Our clients trust us with crucial business details and we will treat that responsibility with the respect it demands
  • We will establish clear guidelines for data governance and update as needed
  • Maintain consistent visibility and control over risk mitigation, ethical, security, and quality standards
Principles

Guided by our principles

These principles anchor us, ensuring AI amplifies our impact without compromising our standards.

Maintaining the same standards as before

Experimentation never comes at the expense of trust. Through this next wave of technology, we remain good stewards of our clients' businesses, data, and reputations.

Data protection comes first. We never input client specific or sensitive internal data into AI tools unless the tool is Crema approved, we have explicit stakeholder consent, and data is properly secured.We own the outcomes.

AI may assist, but Crema signs the work. Teams remain responsible for quality, accuracy, and ethical alignment. Final review and approval is always human led.

Everything is transparent to clients. We're open about when and how we use AI, and we love exploring the possibilities together. Intellectual property, user experiences, and critical business decisions are handled with care.

And that’s a promise.

Our UX Audit Process

1. Understand
2. Analyze
3. Synthesize

To maintain a streamlined and efficient communication channel, we schedule weekly Q&A sessions. This regular interaction ensures that all parties are aligned and can quickly address any queries or adjustments needed as the project progresses.

Our UX Audit Process

Understand
Gain a baseline understanding of your business context
  • Clearly outline the business goals
  • Evaluate the competitive landscape
  • Determine usability and compliance needs
  • Review prior research
01.
Analyze
Gain a baseline understanding of your business context
  • Honeycomb framework
  • SWOT analysis
  • Examine user flows and content quality
  • Validate findings with direct user feedback
01.
Synthesize
Summarize findings and outline actionable recommendations
  • Prioritize suggestions on impact and feasibility
  • Provide a clear roadmap for enhancing UX
  • Determine next steps to achieve business outcomes
01.
To maintain a streamlined and efficient communication channel, we schedule weekly Q&A sessions. This regular interaction ensures that all parties are aligned and can quickly address any queries or adjustments needed as the project progresses.
Two developers talking and smiling over a computer in a bright garage office spaceA product strategist outlining key points on a whiteboard in front of a co-worker

Bright partnerships ahead

This work is too important and our experience too valuable to wholly delegate to general purpose models. Rather than be led purely by AI, we use it as a lever for our own defining contributions. AI won’t dictate our direction, but help us pursue it more clearly and quickly.

At this current phase, our goal is to experiment rapidly, understand the tradeoffs, prioritize what matters, and deliver work better. The reality is that nothing is set in stone with everything changing so quickly.

We invite you to explore this frontier with us.
It’s going to be a worthwhile adventure.

Get in touch

Common questions around how we use AI tools

How do you know what you should use AI for?

We use the SPARK framework to figure this out quickly. Does your project involve Scale (high volume work), Pattern (repeatable tasks), Ambiguity (needs fresh perspective), Redundancy (been done before), or Knots (team is blocked)? If yes to any of these, AI might help. We test it with a small experiment before committing. Not everything needs AI. Sometimes the straightforward approach is faster and better. We're looking for actual impact, not using AI for its own sake. The real question isn't whether AI could work. It's whether it solves your specific problem better than other options. We help our clients figure that out through focused experiments that show results quickly.

How do you spot AI hallucinations?

AI makes up stuff when it doesn't know the answer. It sounds confident either way. We verify anything that matters. Run the code, check the sources, test the logic against what we already know. If something feels off to someone with real experience, it probably is. We ask for sources to dig deeper and cross reference. Generally we don't publish, ship, or recommend anything we haven't validated ourselves.

How do you find the right AI tool when there are so many options?

Finding the right tool takes experimentation. We run small tests internally before bringing anything to client work. Does it actually solve the problem? Is the output quality consistent? Does it integrate with existing workflows? The tool itself is only part of it. There's a learning curve for your team and change management if you want adoption at scale. A powerful tool nobody uses doesn't help. We focus on tools that prove themselves through repeated use. Not the ones with the best marketing, but the ones that actually deliver when it matters. That list evolves as capabilities improve and new options emerge.

Related content

A Crema engagement is a partnership between your team and ours.

Get in touch to learn more about our services. We'd love to hear about your project.