Building public surfaces carefully

Every institution that builds things in public faces a decision that is easier to ignore than to resolve: what exactly is the public surface, and who is it for?

This is not a communications question. It is an architectural one. The answer shapes what you write, what you keep internal, what you let clients see before they sign anything, and what you protect not out of secrecy but out of discipline. Getting it wrong in either direction has costs. Say too little and you appear opaque: the kind of organisation that exists only in references and introductions. Say too much and you turn proprietary thinking into public scaffolding that competitors can walk straight onto.

We have been working through this question seriously over the past months. It deserves a direct treatment.

The two failure modes

The first failure mode is the organisation that over-publishes. Typically this comes from a desire to appear credible: to signal sophistication by displaying internal complexity. You see it in companies that post detailed architecture diagrams before they have launched anything, or publish roadmaps that read more like investor pitch decks than working plans. The intent is trust. The effect is usually the opposite: a reader who knows the domain can see immediately that the published detail is aspirational rather than operational, which raises rather than lowers their scepticism.

The second failure mode is the organisation that under-publishes. Everything is NDA-gated. Nothing is accessible without a conversation. The theory is that scarcity creates value. In practice it creates friction, and friction at the awareness stage is a tax on every relationship you might have had. The organisation becomes legible only through personal introductions, which caps its reach at the social graph of its founders.

Neither of these is a coherent strategy. They are both ways of avoiding the harder question: what does a thoughtful, permanent public record actually look like for an organisation building sensitive commercial systems?

What the word "public" actually means here

It helps to be precise. "Public" in this context does not mean "visible to everyone." It means available without a gate: searchable, persistent, not requiring a relationship to access. Something published here will be read by people we have never met and will never meet. It will be indexed. It may be quoted. It will outlast the circumstances that prompted it.

That changes how it should be written. A note shared in a client meeting can be provisional, exploratory, almost conversational. A published post cannot be. Not because we need to perform certainty, but because anything put into permanent record should be worth the permanence. It should say something that is true now and likely still coherent in three or five years.

The practical consequence is that this archive will be sparse by design. We will publish when there is something to say that earns its public form. Not to maintain a content cadence, not to signal activity, not to feed an algorithm. Frequency is not a metric we are optimising for here.

What belongs in public view

Some things are obviously suited to the public surface. The orientation of the work, what Mustard Seed Group is building toward, what principles shape the portfolio, how we think about capability and intelligence, belongs here. These are not implementation details. They are the institutional worldview, and a worldview that cannot be stated publicly is probably not as coherent as it needs to be.

Commentary on the external environment also belongs here. When a meaningful paper is published, when a technology shifts the landscape, when something in the broader AI research community changes what is plausible: this is worth recording with our interpretation attached. Not to demonstrate that we are paying attention, but because our reading of these shifts is part of the intellectual record we are building.

What we are genuinely learning is also appropriate for the public record, provided the learning is stated at the level of principle rather than the level of implementation. There is a meaningful difference between writing "we have found that scoring commercial conversations by intent quality rather than volume changes which opportunities we pursue" and writing "here is our lead scoring algorithm." The first is a principle. It may be useful to others. It cannot be directly lifted. The second is an asset, and assets that are commercially relevant should not be handed over for free.

What should stay internal

This requires less explanation but more discipline to actually maintain.

Private: specific system designs, the logic inside any of our scoring or ranking systems, the structure of how we handle client data, anything that amounts to a methodology a competitor could extract and deploy. Private: client relationships, engagement terms, anything that has been shared in a working context. Private: the internal tensions: the decisions that went the other way, the approaches that did not work, the strategic arguments that are still live.

That last category is the interesting one. There is a genre of founder writing that performs vulnerability by publishing internal uncertainty. "We almost shut down. Here is what we learned." This is often valuable when it is retrospective and when the organisation has genuinely moved through the experience. It is almost never appropriate in real time, and it is never appropriate when the uncertainty involves clients or commercial relationships.

The test we use: if something could be extracted from this post and used against us, commercially, legally, reputationally, it is not ready for the public surface. If something could be extracted and used to build something without us, and that something is a meaningful part of how we create value, it is not ready for the public surface. If neither of those conditions holds, the question becomes whether it is actually worth saying.

The trust question

Publishing carefully is not primarily about protection. It is about trust.

Trust in the context of an organisation like MSG is built slowly and through consistency. Clients who engage with TUXX for commercial AI work need to believe that what is shared in a working context stays in a working context. Clients who use Orbit need to believe their operational data is not being used as material for public commentary. Partners at any level need to believe that working with us does not mean being written about.

The public record we are building here is therefore, in part, a signal. It demonstrates a particular kind of discipline: the willingness to have a genuine public voice without mistaking the public record for an internal communication channel. These are different registers and conflating them is a failure of judgement, not just strategy.

We have seen institutions that lack this discipline. The damage is usually slow: a client who notices that something from a working conversation appeared in a blog post, a partner who realises the relationship is being mined for content. The damage is also usually terminal. Trust is harder to rebuild than to maintain.

On the shape of this archive

This site, mustardseed.group, will over time carry a meaningful body of writing. What we are doing here is not marketing. It is not content strategy. It is the institutional memory of an organisation that intends to last.

The writing will cover how we think about AI capability, what the research landscape looks like from where we sit, how the portfolio is oriented, what we are learning from the work. It will not cover who we are working with, what we are charging, how our systems are built, or what decisions we are still making.

That boundary is deliberate. It is not arbitrary modesty. It reflects the belief that the most valuable thing we can contribute to a public record is our thinking, the orientation, the analysis, the principles, rather than our methods or our relationships.

The methods are how we create value commercially. They should be earned, not gifted.

The relationships are not ours to write about.

What remains is the thinking. That, we can share. And over time, if the thinking is good, it will be the thing that makes this archive worth reading: not because it reveals how we work, but because it demonstrates that we have worked hard to think clearly.

That is the public surface we are building. Carefully, deliberately, and with the intention that everything here will still be something we would stand behind in five years.