AI Norms & Values, Part 3 of 3: Things We Hold True

Welcome to the third and final part of our series on AI norms and values. Parts of this doc were extracted and published separately on substack; as a whole, they describe the principles we hold pertaining to technology and AI, and the ethical commitments we make to each other and our customers.

We set out to write about AI, and ended up writing about ourselves. These documents are not meant to be aspirational ones; they are derived from how we do our work every day in honeycomb.

This concludes the series, but not the discussion. Dr. Cat Hicks, author of The Psychology of Software Teams (out last month!), has contributed greatly to the industry's research and reasoning around how to build an excellent, high-performing, and deeply humane environment. Dr. Hicks and I will be continuing the conversation on our respective blogs over the next few weeks. We will also do a new episode of Leading With Observability and an open invite webinar where you can bring us your gnarliest questions about how to be a human in the AI era.

This is a weird time to be in computing. So much has changed, so fast. As a technologist, it's hard not to be dazzled by the possibilities, the speed, the range. As a human being, it's hard not to be tired.

On one hand, AI is just technology. On the other hand, this technology is different. Instead of the brute force of automation, AI presents an uncanny valley, a smooth facsimile of cooperation and cheer. I think this accounts for the creeping suspicion that we are not being treated like real people. But AI did not invent being rude, disrespectful, or careless with other people's time. For every problem amplified by AI, there are solutions AI can accelerate.

AI is not the point. The point is us.

Tools are just tools. An email can bring people together or tear them apart. AI can be used to avoid other people or to build and enrich communities. The difference consists of intent, understanding, consent, and good fit, not technology.

We are a company founded on values and first principles, and the stubborn conviction that technology can be so much better than people are used to. We're still doing that… just with AI.

Things we hold true

AI is a tool

AI is a powerful tool, but it is just a tool. We do not serve our tools. Our tools serve us.

You own your work

I am not a “human in the loop,” I am the owner of the loop. It's my fucking loop. I am responsible for the quality of my work and the integrity of my working process, and so are you. “Claude did it” or “Claude said” is not an excuse.

The bar is going up

Every technological transition resets the baseline for performance—not by mandate, but by what becomes possible. This is true for companies, products, teams, tools, and people. What was excellent five years ago is table stakes today.

We welcome this. The nature of technology is that it always catches up. Which is why the nature of technologists is, we stay ahead.

The outcome is what matters

A higher bar means better outcomes. Often this is about moving faster, but not always, and never exclusively. What outcome are we aiming to achieve? What would ‘better’ look like? What would ‘great’ look like? What is possible today that wasn't possible last year?

For outcomes to matter, we must embrace measuring ourselves. Everything is an experiment, but an experiment that doesn't get tracked is wasting everyone's time.

How we use AI

A shortcut or a deep dive

You can use AI to help you do something fast and half-assed, or you can use AI to help you think harder, build with more rigor, communicate with more depth.

There is a time and a place for both, but the two are not interchangeable. Know the difference.

AI for the role, AI for the self

You may use AI at work in ways that are prescribed by the company to do your job, and we owe you enablement and support for those.

You may also use AI of your own volition for calendar support or editing, as a tutor, rubber ducky, etc. This is much more personal and subjective, and no one is required to do it, but opting out of using AI does not exempt us from the higher bar.

More AI is not always better

AI is not the right tool for every use case, and more AI is not always better. If AI is causing friction and frustration in a given setting, talk with your team about how to reduce or eliminate that frustration. Make the tools serve you.

Master the tools before ruling them out

However, be careful to make this decision from a place of fluency, not ignorance, and revisit your decision occasionally. It's easy to blame the tools for the frustration of learning.

Usage patterns

AI should make our signal-to-noise ratio better, not worse. To do this, we need three things: self-awareness, respect for each other's time, and open dialogue.

Address the envelope

Any time you send someone an artifact, tell them what it is, why you're sending it, and what you hope to receive from them, and any other useful context you can think of.

This is the “envelope” of your message. Addressing the envelope does a couple vital jobs:

• It cultivates self-awareness. When you drop a large artifact on someone and walk away with no comment, it's easy to breeze past the fact that you may have just dumped a ton of work on someone. This surfaces as frustration with AI because AI has newly made this easier, but AI is not at fault. You own your work. When you describe your request in detail, it forces you to think through what you are asking them to do and what you really need from them.
• It opens a conversation. When someone drops a link on you, it can feel like a demand, but it's not. We don't assign work at Honeycomb, we ask nicely, and if someone has concerns or objections, we talk about them. This should always be a conversation. “This is what I have, this is what I need. Are you the right person? Is this the right time? What questions do you have?” Ask. If someone sends you an artifact or a request and you feel a pit in your stomach, pay attention to it. What is it you fear?

Respect other people's time and attention

Never send someone an AI-generated doc unless you have first read it yourself, every word. If it's not worth your time and attention, how can you possibly say it is worth someone else's? This may be a low bar, but it's an important one.

To respect other people's time and attention, make the smallest necessary ask. You wouldn't ask someone to read a whole book when you need them to read page 152. Specificity is kindness, boundaries are respect.

Ethical issues we have a stance on

There are two specific mistakes we are trying to avoid in writing this section.

We don't want to write something glossy and aspirational about how much we “Care About Ethics™” with no evidence to back it up, but we also don't want to make commitments we can't keep. Instead, we will affirm some of the principles that have gotten us this far, and describe how they have guided our decisions in the past.

Externalities and harms done by AI

There are a number of troubling externalities associated with AI:

• Energy usage/carbon footprint
• IP theft
• Privacy
• Bias
• Job cuts, wage cuts, and disinvestment in the next generation

We are a for-profit company, and our first priority is to build a successful, sustainable business. When we succeed at business, we earn the right to think longer term and make bigger investments.

Our vendor review process includes an ethics review. We preferentially give our money to vendors that share our values or are the lesser of two evils. We will continue to do so.

On energy and resource usage

We frankly have no idea what to do about this, as a consumer (not producer) of foundational AI models. Which is unfortunate, since it is probably the most alarming externality. We will re-examine in six months.

On intellectual property

This is a thorny one. It appears that model providers have knowingly violated copyright law for years with no brakes and no consequences, outcompeting (and dooming) competitors who followed the law. Was it illegal? Sure seemed that way. But the only honest answer is we don't know, and we won't know, until we learn how the Supreme Court will interpret the law.

But was it right? Absolutely not. It was settled law at the time, and they systematically broke it. To retroactively bless this behavior sets a concerning precedent for businesses like ours who believe in fair competition and following the law.

Ethics, morality, and the law have always been different things. As model users, we are aligned with the letter and spirit of the laws. But we benefit from the unsavory acquisition of training data, and we know it, which incurs a moral debt.

At a minimum, we should be diligent in citing references, assigning credit, and compensating contributors whose work we use in commercial activity, above and beyond what is legally required of us. All of which we have a track record of doing, and will continue to do.

On bias and privacy

“You own your work” means that any bias introduced by AI is your problem. Be mindful of where bias tends to creep in, and guard against it. Verify results where possible. Welcome feedback warmly, correct, and move on.

It is not clear how we can contribute to privacy efforts.

On wages and disinvestment

We have never been a company that believes in squeezing the most work out of people for the least money. We compensate as well as we are able, and we take fair pay very seriously.

We have always felt a responsibility to invest in the people who work here, just as they invest in us. This includes hiring and training entry level workers when we are able to do so.

Mentoring and learning are at the heart of our job ladders in R&D, and the SDR pipeline serves a similar purpose on the GTM side. We have supported a number of employees in retraining for different roles. Personal development budgets are widely used. We audit our salary bands for bias and equity, and practice transparency around pay bands in job postings.

We have learned the hard way that it does not serve anyone for us to hire junior employees when we aren't ready to support them, but when we can support them, we have and we will.

On activism vs working agreements

These ethical stances may seem relatively modest, and they are. These are working agreements for doing business, not ethical aspirations or activist goals.

There is a place for activism. There is a place for outright advocacy. At a time like this, we would all be well served to consider our ethical stances and how to act on them. Business may not be the ideal vehicle for activism, but activism is vital and necessary. People who take action in support of their beliefs (generally within the confines of the law) will not be retaliated against at Honeycomb.

We are here to build a business. There is virtue in this, even if it is not explicitly ideological.

Closing statement

The workplace is one of the last remaining places where people of widely varying backgrounds and beliefs all come together to achieve something greater than themselves. We think this is precious. We think this is worth protecting.

We believe that treating people well is not at odds with the profit motive. We believe that people who are happy, healthy, well supported, and creatively engaged can do better work.

We believe that rigorous use of AI helps a company like ours accelerate development, delight customers, and go head to head with competitors with vastly more resources. We believe that using AI is not a replacement for skill and craft, but an amplification.

We believe that customers who are valued, respected, and listened to are happier, more loyal customers. Happy customers are more invested in giving us the feedback we need to build a better product. We believe in building mutually beneficial relationships and positive feedback loops that leave both sides better off.

We do these things, not as a sacrifice or a distraction from our core mission, but in service of it.