Information Integrity

I’ve been using AI agents for writing code heavily over the past few months. I step in and make small adjustments here and there, but I’m challenging myself to avoid writing code by hand. I diligently review every output, but I let the agent write the code. It’s been an eye-opening experience to see what the future of software development looks like, especially knowing that, right now, the models are the worst they will be. They are more capable with each iteration, so knowing how they work, how to use them, and, most importantly, how to work with them is critical.

Working with code in this way made me think about how information moves through the work, like an invisible fiber weaving context into clarity into value. Because that’s really what product work is: moving information through in a way that maintains its integrity and transforms into realized value in the final product. When work isn’t delivered or misses the mark, it’s because somewhere along the way information degraded. Context wasn’t shared; intent wasn’t understood; assumptions were made.

The result? Quality drifts and value degrades.

At key points, information turns into either:

• Quality Drift when information degrades and loses some of its value.
• Compounded Value when information sharpens and increases its value.

Working with AI agents to write code, you provide instructions as the source of information, and the model writes code and attempts to translate your intent into form.

This is a microcosm of the product process.

You collect data from users, stakeholders, and the market; you translate that into a problem to solve and solutions to address it; you translate that into designs and technical specifications; you translate that into working code in your product; you collect feedback and translate it more problems and solutions.

Translate, translate, translate.

Product work is translating value, taking raw data and transforming it into an impactful return on the time you spent getting there. It’s a simple equation of inputs and outputs. Your success depends on your ability to uphold the integrity of information through the lifecycle. And the more languages you speak—sales, marketing, design, engineering, strategy—the better equipped you are for the translation work. This is why generalists and Product Builders thrive: they have high fluency and heightened ability to maintain the quality of information from concept to creation.

Now, there are two ways to look at information as it pertains to AI coding agents:

1. They add more translation points and increase risk of quality drift.
2. They decrease translation points and increase the compounded value.

In the former, you’re following the standard process outlined above. And, instead of an engineer directly translating a shaped definition in code, they’re prompting an AI coding agent to do it, adding a new layer of abstraction—and another translation point. It’s not necessarily bad, but it is another step at which information can degrade.

In the latter, you’re reducing upstream translation points and working with higher speed and output to reduce risk. Because, if you think about it, what’s the real reason so much product work is about planning, prioritizing, and preparing the work? Because of the downstream risks involved in the uncertainty of writing code with costly engineers. And even after all that planning, prioritizing, and preparation, you can still miss the objective with delayed timelines, low-quality features, or bad bets. That’s why software development is so broken.

But there’s a better way.

AI is a part of the seismic shift in the paradigm. And you can use it as a collaborator and accelerator while maintaining the craft, the quality, and the coherence of your product.

This idea is called Information Integrity, which lives in the Clarity Current of the Claritorium and Value Creation of Equilio.

The three pillars are:

1. Articulation to transform intent into shared meaning.
2. Translation to transform shared meaning into structured form.
3. Realization to transform structured form into realized value in the world.

The Formula

The formula for Information Integrity can be expressed as: Event + Mode = Outcome

The Event is represented by the pillars of Articulation, Translation, and Realization.

The Mode is either:

1. Dictation when a one-sided message communicates information.
2. Integration when multiple parties engage in a collaborative dialogue.

And the Outcome is one of the pathways we met earlier: Quality Drift or Compounded Value.

Articulation

Articulation is the act of framing through distinct language—an expression of meaning.

Humans developed complex language to imbue meaning and support communication toward shared efforts. Language is a tool, and, like any tool, there’s skill involved in its proper usage. As I write these words, I’m choosing specific words to communicate an idea. I’m articulating information in the hopes of creating shared knowledge more broadly. Whether I succeed or not isn’t up to me. You get to decide whether I articulated the words in the right sequence and cadence to maintain the integrity of information.

Too many words and information can distort.

Too few words and information can be lost.

When talking to a customer in an interview, how you ask the questions matters. Are you leading them or letting them lead you?

When you cast vision to your team, the right words make or break whether it sticks. Are you compressing the idea without losing quality?

And when you write prompts for AI models, the quality of context is even more important. Are you clear with your intent, or do you expect the LLM to fill in the gaps? Because, despite their robust confidence, they’re mere prediction engines. Like humans, they require sufficient context to understand. Unlike humans, their knowledge bank is limited to the information they’ve been exposed to, devoid of human experience and nuance. Talking to Claude Code is a different mental model than talking to one of your senior engineers who holds deep tribal knowledge with their own unique experience building software.

To maintain the integrity of the information, you need to operate in the right mode. Don’t just dictate information:

Fix the bug in signup flow.

Try it. I suspect the LLM will confidently do something, but I guarantee what comes out the other side will be insufficient. It’s guessing. A human with the right knowledge and experience would make a better effort.

Try something else: integration. Talk with the LLM like you’re starting a dialogue, not passing off one-way instructions. Great ideas and solutions emerge in the symbiosis of conversation; multiple parties sharing ideas and building a beautiful alchemy of creation.

What if we said this instead:

There’s a bug in the signup flow. We made a change to detect the user’s authentication method based on their email address, but when we can’t find them in our database, the signup type is coming back as an error, which prevents them from signing up. Look at the implementation and report your findings. Ask questions and suggest focused changes to solve the problem without creating new issues. We can’t fix this and break something else. View the change holistically as part of the entire authentication flow. We should also add a specific test case so this doesn’t happen again.

Why does this matter? Shared meaning.

Without information communicated through clear articulation, meaning is lost. And the only way you know if shared meaning is established is through dialogue—through the integration of sharing information. I don’t actually know if what I’m expressing here resonates unless you, the reader, say so. Most of these ideas expressed here surface through work and interpersonal conversations. The writing is for me to create shared meaning, yes, but I’ll never know what’s truly shared without the dialogue.

If you want to uphold the integrity of information, your first step is articulation. Find the right language through interaction, dialogue, conversation. Mix ideas, opinions, and rigorous debate into a wonderful medley of shared meaning that resonates. Whether you are working with your team or LLMs, articulation is the gateway to clarity.

Articulation is an exercise in reducing information into its simplest form without losing its quality; compression without distortion; signal without noise. When working with people, the simpler, the better. When working with AI, that’s not the case. You need to be clear, but you need to provide enough context to form the foundation. But once you’re deep in the conversation, or there’s sufficient enough context loaded in, then you can concentrate your message. The key is to strike a balance, and know when articulation requires expansion or compression, controlling the surface area of information.

Principles

1. Keep it simple. Compress the idea as simply as you can without losing the meaning. Make it clear and memorable.
2. Find the analogy. The easiest way to understand something is to see how it relates to what you already understand.
3. Make it visual. A picture forms a visual mental model that’s easier to interpret and understand as a representation of meaning.

Practices

1. Idea Compression: Spend time reducing the information into the smallest atomic unit of understanding and meaning.
2. Meme Generation: Think in memes as representations of ideas that spread. Turn the information into its own meme.
3. Language Socializing: Words have a different feel out in the open. Information improves as you think out loud and see how it resonates with others. Talk it out.

Translation

While Articulation is about creating shared meaning, Translation takes shared meaning and converts it into a structured form.

The framed problem becomes an opportunity.

The opportunity becomes a design.

The design becomes code and function.

The product generates user feedback.

The feedback fuels new problems.

Information weaves through a labyrinthine of discovery, design, development; framing, shaping, building; discussion, commitment, iteration; unknowns, assumptions, risks.

Throughout the journey, information requires a shepherd to nurture and protect it.

But information can and should change. Ideas don’t stay exactly the same from the initial idea all the way to the end product. You don’t want to lose the essence of the idea, which is why translation work is so important. Most software development is bloated with endless ceremony. If the risk of quality drift is proportional to the number of steps involved in the process, it’s obvious why so many half-baked features end up in products. And, through the process, information is handed off from team to team, person to person. Or, even worse, no thought is given to the request and it’s implemented as stated, no translation.

Information is shaped by constraints. This is the primary driver of translation work because each step brings new constraints to consider: time, money, resources, feasibility, usability, skills, experience, knowledge. But what happens when AI makes it easier than ever to convert an input into an output? If the right person can leverage an LLM to translate an idea into a product, how does this all change?

Product Builders have the right skills, amplified by AI, to reduce handoffs in the typical software development process. I feel like I can do the work of 10 engineers with tools like GitHub Copilot, Claude Code, and Linear with its agent integrations. It’s like having an always-on expert ready to respond immediately. But you still need to know design, engineering, and strategy to use it effectively. You have to review the output, provide guidance, and think critically while you engage in a dialogue. Working with AI should feel like you’re solving problems together, collaboratively. If you just dictate and ignore layers beyond the surface changes, you lose control of the holistic quality. You build a house with nicely painted walls held together with silly string.

Translation turns shared meaning into structured form through constraints as representations: flows, UI states, system behavior, data models, edge cases, tradeoffs.

As the shepherd of Information Integrity, your job is to maintain coherence across the representations. And the best way to do that is through a deeply integrated understanding of multiple perspectives and experiences. Only then can you become fluent.

Principles

1. Let constraints refine the solution. Constraints are shaping forces to maintain meaning in the structured form. Use them.
2. Name trade-offs before they harden. Quality drifts when you don’t consider and actively address the inherent trade-offs.
3. Optimize the whole, not the part. Local correctness can still create global failure. Look deep and wide as you translate into form.

Practices

1. Constraint Mapping: Dedicate time to explicitly calling out key constraints like time, feasibility, usability, and anything else before committing.
2. Decision Log: Write down trade-off decisions so you can understand what was changed, why, and how to learn from the decision.
3. Shaping Sessions: Bring together design, engineering, and strategy to shape forms in low-fidelity that blend each perspective into a unified whole.

Realization

The culminating event of Information Integrity is Realization. This is where the structured form created through Translation lives in the final context.

If Articulation is asking do we mean the same thing?

And Translation is asking can that meaning survive constraint?

Then Realization is asking does the structured form survive reality and create real value?

Realization is where structure meets reality.

Does the solution truly solve the problem?

Does the customer know how to use the design?

Does the code powering the product hold up under load?

Everything up until this point is a simulation, a proxy, an attempted representation of the product in the living context. When Information Integrity is upheld and the meaning is shepherded through successfully, value is compounded. If, however, information is poorly articulated, meaning is lost, and the structured form is misrepresented, then quality drifts into an unusable mess. All the days, weeks, and months of work in vain.

When you’re articulating information, it’s important to define intent:

What does success look like?

When you work through constraints, you need an anchor. The definition of success is your anchor to maintain Information Integrity, measure impact, and deliver with confidence.

We just released a redesigned sidebar in our AI chat experience. I worked with one of our designers to get enough of the structured form in Figma before moving to code. I used GitHub Copilot and Claude Code to work off of the mockups and build the prototype. I gave it a long prompt with some standard boilerplate I give any agent before a task. We also have custom agent Markdown files in the codebase to give context about the project.

It did well on the first pass, but there were many iterations to work through. I reviewed and modified the design, but also the code. I told it how to combine and architect the components, what tests to write, and where to borrow from existing styles. It’s faster for AI agents to create new code than to review and pull from existing styles. You have to instruct it to do so. And, in the process, I wasn’t just reviewing the surface changes; I was also looking at the code because it’s part of the entire solution. I was engaged in a dialogue, an integrated workflow with AI.

I ran into unexpected side-effects with the change. Other parts of the interface felt out of balance as a result of a new sidebar. We wanted to keep the time constraint, so I decided to make minimal changes to balance things out and ship it. The constraint gave clarity. And working in the real environment is where the form transformed to value. Moving faster (with AI) to get feedback from real users is the only way to truly measure impact.

Realization comes from reality.

Principles

1. Reality is the final authority. The faster you can get the information to live in the real world, the faster you validate its value.
2. Durability matters more than delivery. Value is measured over time, not immediately. You need to keep with it and refine against feedback.
3. Learning is the measure of success. If you don’t learn something from what’s delivered, then you didn’t success. Even failure is learning.

Practices

1. Holistic Reviews: When working with AI, don’t just review the feature; review the code, the second-order effects, and all the other integration points.
2. Variable Scope: If you want to deliver on time, think about how you can change the scope and still solve the core problem, the essence.
3. Reality Checks: Compare intended success criteria to actual outcomes and understand the delta. Keep refining to go in the right direction.

The Throughline

Articulation protects meaning; Translation protects coherence; Realization protects durability.

Each event is a choice to:

1. Maintain Information Integrity and compound value;
2. Or let quality drift and waste time, energy, and money.

AI—and using AI agents to write code, specifically—is challenging every assumption of software development. It can be scary, yes, but there’s also a huge opportunity to move work into reality faster, learn, and deliver more value. Iteration speed in action.

I’m having the most fun building software right now. I still choose to engage with the output, collaborate in its creation, and create more space to do the work I enjoy most. Information is the raw material of creation. But only when you maintain its integrity from concept to creation.

So, will you be an architect of change, of value, of creation?