A connected ecosystem

What makes a portfolio a portfolio

There is a particular failure mode in holding companies that try to span multiple categories: the work looks diversified from the outside but is actually disconnected from the inside. Each product operates on its own logic, builds its own conventions, hires to its own culture. The centre is administrative, not intellectual. The companies happen to share a parent rather than a thesis.

Mustard Seed Group has been thinking carefully about how to avoid that.

Not every tension between product areas needs to be resolved. Consumer and commercial are genuinely different disciplines, and it would be a mistake to pretend otherwise. Fitness coaching is not the same problem as business operating systems. Research infrastructure is not the same as a client services studio. These things should feel different because they are different.

But underneath the different surfaces, there are questions that run across all of it. What does good execution look like at the level of a person? How does an organisation stay coherent under complexity? What is the right relationship between a human and an intelligent system? When should a product automate, and when should it leave the decision with the person?

Those questions belong to the whole group, not to any one product.

The commercial layer: Orbit and Orion

Orbit has been growing into a clearer shape over the past year. The idea, a business operating system for the full lead-to-launched-product workflow, is not a common one. Most commercial software carves out a single stage and goes deep there. CRM covers relationships. Project management covers tasks. Finance tools cover numbers. Each category assumes you will handle the joins yourself.

Orbit does not accept that assumption. The argument is that the joins are where most commercial execution actually fails. A lead gets qualified but never converted to a proposal. A proposal gets signed but the work never gets properly scoped. A scope gets built but delivery drifts because there is no structured handoff. The problem is not that teams lack good tools in each stage: the stages do not talk to each other.

That is a product design problem, not a workflow problem. You cannot solve it by choosing better software for each compartment. You solve it by designing the surface around the full arc.

Orion is what makes that arc intelligent rather than just structured. The intelligence layer underneath Orbit is responsible for the work that structures alone cannot do: remembering context across long commercial cycles, surfacing the right information at the right moment, reasoning across relationships and stages, and helping teams know what needs attention without having to ask.

The relationship between Orbit and Orion is worth being clear about here, because it is not simply a product and its AI feature. Orion is a research project as much as an engineering one. The questions it is asking, about memory, context, orchestration, the right level of autonomy at each stage of a commercial workflow, are not questions that have clean answers yet. Building Orion inside a real product forces the research to stay honest. Abstract intelligence research is easy to overclaim. Orion cannot claim success until an Orbit user can actually move faster and more accurately through real commercial work because of it.

The services layer: TUXX

TUXX occupies a different position in the group. It is the part that works directly inside client environments, builds custom systems to specification, and proves patterns under real-world constraints.

That last phrase matters. Benediction Lab can theorise. Orbit can design. But TUXX has to make things work where real people are trying to do real jobs, with existing tools, existing habits and real deadlines. That is a harder test, and passing it teaches things that no amount of lab research can substitute for.

Pattern Up, the TUXX sub-product, is an attempt to formalise some of that learning. The patterns that appear repeatedly in client delivery, the shape of a brief, the structure of a review, the point at which a workflow needs a human decision rather than an automated one, those can become templates, tools, a repeatable surface. Not every client needs a fully bespoke system. Some need the repeatable version of patterns that have already been proven.

The value TUXX creates for the wider group is not just revenue. It is signal. When something breaks in a client system, that break reveals a real gap. When clients adopt a pattern quickly and enthusiastically, that adoption tells us something is worth productising. Services, done well, are one of the most honest research methods available.

The consumer layer: CheekyGains and All Purpose

The consumer side of the portfolio operates on different logic. The audience is not a buying team trying to close a deal. It is an individual trying to become a better version of themselves.

CheekyGains and All Purpose share that orientation, though they express it differently. All Purpose is the broader ecosystem: creative culture, music, community, the practices that make up a creative life. CheekyGains is the performance and fitness expression of it: accountability, standards, coaching, the work that goes into showing up physically.

Naira, the AI coach inside CheekyGains, is one of the more technically interesting ideas in the consumer portfolio. The temptation with AI coaching is to make it generic: broad enough to serve everyone, specific enough to feel personal, but without the depth to actually shift behaviour. That is not what Naira is being built to do.

The harder version of AI coaching is one that understands standards. Not just "you said your goal is three workouts a week, you did two". That is a tracker. A coach understands why two became two, what the pattern behind it says about the person, what the right response to that pattern is, and how to deliver that response in a way that actually lands. That requires specificity about the individual, not just their statistics.

This is a hard design problem. It requires the same serious thinking about memory, context and the right level of automation that Orion is working through on the commercial side. The domain is different, but the structural challenge, building intelligence that is genuinely helpful rather than impressive, is recognisably the same.

The research layer: Benediction Lab

Benediction Lab's job is to stay further ahead than any single product needs to be today.

The research areas that the Lab has been working through, agent systems, memory architecture, graphical interface control, autonomous product development, are not being pursued because they have immediate product applications. They are being pursued because they are likely to become important, and building understanding before they are urgent is how you avoid being permanently reactive.

The field moves fast enough that waiting until a capability is proven before researching it is already too late to build deep competence. The Lab provides the group with a longer time horizon than any individual product can afford to operate on.

The relationship between Benediction Lab and the rest of the portfolio is not a pipeline where research inputs become product outputs on a predictable schedule. It is more like a library that the rest of the group can draw from. Sometimes a piece of Lab work becomes directly relevant to an Orbit feature. Sometimes it shapes how TUXX approaches a client problem. Sometimes it sits for months before its application becomes obvious.

What the Lab cannot be is untethered from the group's actual thesis. Research for its own sake is interesting but not the mandate here. The mandate is research that eventually makes the products more capable or protects the group from being blindsided by a capability shift.

The connection that matters most

If you look at the six areas, Orbit, Orion, TUXX, CheekyGains, All Purpose, Benediction Lab, and ask what connects them, the obvious answer is intelligence infrastructure. They all use models, they all think about agents, they all care about what memory systems should look like at the product level.

That is true, but it is not the deepest connection.

The deepest connection is a shared answer to a harder question: what should an intelligent system actually do for a person?

The wrong answer, and it is tempting, is to do as much as possible. Automate everything automatable. Generate everything generatable. Optimise every workflow that can be expressed as an optimisation problem. This answer produces products that are technically impressive and practically disorienting. People do not know what the system has changed, what it is responsible for, or what they should still do themselves.

The right answer requires restraint. A system that increases human capability is not one that replaces human effort: it is one that directs human effort more accurately. It removes the friction between what a person intends and what they are able to execute. It does not substitute for intention. It amplifies it.

Orbit should help a commercial team execute their strategy, not generate one for them. Naira should help a person train with more clarity, not tell them what their goals should be. Orion should help a team see what needs attention, not manage their relationships on their behalf.

That is a design philosophy. It has to hold across all of it.

The note for November

What the portfolio looks like in November 2024 is a group of companies that are distinct in audience and execution, but legible as a single project when you look at the level of first principles.

That legibility is not guaranteed. It has to be maintained. It means each product team has to be thinking about the shared thesis, not just their own problem. It means Benediction Lab has to be producing work that is useful to the products, not just interesting to the researchers. It means TUXX has to be surfacing real-world signal, not just delivering projects and moving on.

The work ahead is to keep the connections live. Not to merge the products, not to homogenise the cultures, not to centralise everything into a single system. Just to keep the intellectual thread clear enough that the work reinforces itself rather than pulling in random directions.

A portfolio is only as coherent as the thinking that holds it together. That thinking belongs to the whole group, and maintaining it is one of the permanent responsibilities of running the institution.