Thinking

Sovereign Substrates

Thu, 06 Aug 2026 00:00:00 +0000

AI adoption in the enterprise has been slow, expensive and inconsistent.

Despite significant financial commitment, several organizations have reported exhausting their AI budgets earlier than anticipated, and having to place caps on expenses mid-year without concrete results.

The reasons are many and they point to the organization and not the technology. Enterprise finance needs predictability in costs, and a way to understand ROI.
There are operational mismatches as organizations tend to work in vertical silos whereas AI operates best with lateral access to understand and access cross-functional workflows and data. Metrics for measuring success vary, internal expertise is unevenly distributed, vendor lock-in and data protection risks are real, and with geopolitical entanglement endorsed at the governmental level, there are policy and access risks.

At a higher level there are governance, ownership and budgetary concerns, and this is before model selection, routing and varying API costs.

These bottlenecks are pre-inference; before an AI model is called.

Cheaper models are not the solution

If financial costs were the only obstacle, the availability of excellent and cheaper OSS models from Chinese labs Z.ai, Tencent, Deepseek, Minimax, and Kimi would have circumvented the problem and the current landscape would look very different.
Lower costs, though desirable and advantageous, haven't proven to be the key that unlocks scale in the enterprise.

The Missing Foundation

AI, despite being intelligent, multimodal technology capable of reasoning isn't plug-and-play because it requires inputs and context, permissions and provenance, and clear security and governance controls.

The ecosystem required to make AI work inside the enterprise requires a significant architectural change; a sovereign, pre-inference, model-agnostic layer inside the organizational boundary which augments existing infrastructure, assets and workflows, while protecting proprietary data and processes.

Context is continuously produced by legacy applications, which means either updating those apps, or adding a reliable data conversion layer to their output, so that a machine-readable substrate becomes the organic byproduct of everyday use.

Starting with this, the substrate can be extensible but requires at least three additional layers:

A canonical layer to establish organizational state (what is true now), and memory (what has been learned) including context (what is relevant to the task) and provenance (where it came from)
A governance layer for policy, identity and permissions
An execution layer for model routers, agent runtime and tools etc.

Why This Design

The criterion outlined a year ago in Agents of Chaos is more obvious today.

"Maintaining stateful interactions and memory adds complexity and most companies aren’t ready to invest in the infrastructure needed for real agent-like systems when the technology arc hasn’t reached maturity. Even in the B2B context, it is a reenactment of the Cloud-based vs on-prem debate as organisations require personalisation - knowing their structures, workflows, repositories - for the technology to be truly effective, and this has security and dependency implications."

This encapsulates the struggle enterprises have with integrating AI today: vendor dependency, governance and policy, memory and context, and data security.
As posited, coding agents have continued to improve, arguably more than expected, and the conclusion that true agency requires personalization still holds.
In enterprise terms, personalization is a durable, governed understanding of owned and protected data, structures, workflows and repositories.

Separating knowledge from governance, policy and execution gives organizations control over context, identity and proprietary data prior to inference. Additionally, prefixing data conversion pipelines ensures that used data is current, and that incremental integration efforts minimize friction.

Sovereign Does Not Mean Self-Built

This is likely to operate best as a composable substrate with a mixture of internal and external tools. The effort to build everything in-house would be cost-prohibitive even if it were possible. No organization, even the most sophisticated tools, is capable of creating, maintaining and operating all of its software without diluting its focus.

Sovereignty here means absolute control with interchangeable components making up the less important parts of the stack, while its core is owned and doesn't require the organization to be beholden to any specific vendor for its operations and growth. Most importantly perhaps, the substrate illustrated above lets most human capital continue doing what it does without having to relearn or upskill, while simultaneously infusing the organization with the intelligence and efficiencies of AI.

A composable sovereign substrate is the architectural requirement for seamless AI integration in the enterprise.

Anomaly's Stance: Composable Proof-of-Concept

We believe integrating AI fundamentally changes the operation, value and capability of software, and this requires a new architecture for human-software interaction. Paradox , for example, turns a user's work into a reusable, persistent, machine-operable context layer.

As inference costs collapse, capabilities standardize and intelligence converges at comparable levels, the catalyst differentiating performance for the enterprise class, or any application for that matter, will be the substrate and not the model or the harness. Building this infrastructure inside the organizational boundary is strategically more sound and sustainable than chasing models and re-platforming annually, while being vulnerable to variable costs and data leakage.

To illustrate this principle, we’ve built a narrow, composable component that can slot into any organization's data conversion layer. Skygate is a candidate intelligence data platform to help organizations create structured, machine-readable candidate objects.

Skygate dissolves talent data to make it polymorphic: information imbued with meaning and structure, separated from its original form, machine-operable, and recomposable into whatever output that context demands.

Recruitment firms and large employers can turn resumes and hiring documents into structured, reusable data that can power existing systems and the next generation of AI-ready and operated applications, while keeping data and workflows private, remaining model-agnostic and increasing productivity.

It is upon a foundation with similar traits that hybrid human/machine corporate systems and structures will likely be built.

Computation Is Not Thought

Thu, 05 Jun 2025 00:00:00 +0000

All the talk of reasoning models, agents and LLMs that 'think' may inadvertently set a dangerous precedent; one in which society accepts that AI models are capable of thought, a uniquely human trait.

Familiar Footing.

As humans who witness technology permeate through society and become ingrained in our daily lives, most of us overlook the effect it has on subsequent generations who will grow up as natives in a landscape with that technology already infused in it. This risks a portion of the population having a blind spot to some of the ramifications of technology.

For example, a high percentage of children today would likely meet the threshold for addiction when it comes to digital dependency, and how it shapes priorities, decisions, actions and behaviour, yet strategies to counter it are sparse. This likely stems from the iPhone's twenty-year dominance as an unregulated cognitive extension. Adults today intuitively understand limits and restrictions because our existence predates the iPhone, but younger generations have been affected to a greater extent.

We potentially face a similar problem with LLMs. When technology that computes is used as a substitute for thinking, it has several detrimental effects:

Humans reduce their thinking
Generated content becomes ubiquitous which lowers overall standards
The capacity to think atrophies.
Inertia wins more often
Aspects of memory change

In the world before cell phones, people remembered dozens of phone numbers with ease. That 'skill' and all that it enabled has silently been lost to time. Similarly, the effort we are willing to make has changed with the abundance of instantaneous information. This has a halo effect on individual behaviour; if we can't find something, we settle for an alternative almost immediately. Copy/paste has carved out grey matter from our collective cognition. As the following article from 2019 alluded to, our thinking may be inversely related to our ability to compute.

AEON IDEAS

How Much Can We Afford to Forget, If We Train Machines to Remember?

Civilisations evolve through strategic forgetting of once-vital life skills. But can machines do all our remembering? An exploration of how AI memory partners may be changing what knowledge we retain versus outsource.

April 8, 2019 Read more

Understanding The Difference

Computation is a mechanistic process which is substrate-agnostic and objective, in that it produces outputs from given inputs. Thought, on the other hand, is a cognitive phenomenon, embedded and contextual while being intentional and semantic. It involves a self-referential layer which machines do not have.

Thinking requires identity since it encapsulates multiple levels of continuous judgments.

Cognitive load-bearing is an integral part of intellectual development. Habits formed by technological convenience leave us predisposed to lighter burdens. Where once we might sit with a difficult challenge, allowing ideas to percolate and develop through sustained contemplation and progressive iteration, we now reflexively reach for our devices at the first hint of mental effort or uncertainty; it's easier.

This leads to an insidious pursuit of convenience and instant gratification, fueled by dopamine-hunting interface design on social platforms, compulsive scrolling habits and misplaced priorities. The result leaves us misguided and weakened; with an eroded capacity to sustain focus, think independently and critically, and make sound decisions. This corruption starts at the personal level and cascades through groups, corporations, and institutions

Aspect	Computation	Thought
Syntax vs. Semantics	Operates purely on formal symbols (syntax).	Involves semantic understanding (meaning) and situational context.
Determinism	Follows explicit rules; given the same input, always the same output.	Although it can use rules, thought often includes intuition, heuristics, or even "non‐algorithmic" leaps.
Consciousness	No requirement of awareness or subjective experience.	Typically involves an awareness of oneself processing ideas (metacognition).
Creativity & Flexibility	Can be highly innovative within defined parameters and rules, but operates within the framework and objectives given to it.	Draws from abstract, unrelated domains to transcend existing frameworks.
Embodiment	Abstract process tied to whatever hardware implements it.	Deeply influenced by bodily states, emotions, and sensory feedback.
Intentionality	A computation doesn't care what it's computing.	Thoughts carry imagination, curiosity, priorities and intentions.
Context Sensitivity	Context must be encoded explicitly; otherwise, no "background understanding."	Draws on situational cues, long‐term memory, and broader world knowledge seamlessly.

LLMs perform large-scale statistical inference over learned patterns.
An answer matches your prompt's pattern to similar contexts encountered during training, then retrieves the next-token that statistically fits best. This points to a massive underlying web of learned parameter values and not an explicitly embedded ontology.

Their non-deterministic outputs arise from introduced randomness which can be controlled through temperature, top_k and top_p settings. This is the realm of computation which draws from a probability distribution computed via matrix multiplications.

Dynamic models with web access or continuous learning can retrieve real-time information via web search or updated databases. They may have the ability to learn or correct themselves, but this depends on how information is weighed, leading to a whole host of ethical quandaries. Since most LLMs are fundamentally limited to patterns present in their training corpus, they can't genuinely encounter something genuinely novel and integrate it the way humans do. When humans encounter a new concept, we can:

Connect it to embodied experience
Relate it to emotional or sensory memories
Instantaneously develop bias or affinity based on a multitude of personal factors, and opinions stemming from our perspective, and ultimately identity
Anticipate future implications based on identity and goals
Modify our entire conceptual framework to accommodate it

LLMs have no qualia and can only interpolate within the conceptual space they've already seen. This means their 'creativity' is more accurately defined as innovation within fixed bounds and rules.

If we conflate computing and thinking, we might erroneously lean on computational solutions for problems that require abstract thinking that only humans are capable of. Or, we might undervalue human cognitive contributions that can't be replicated computationally, which has an extraordinarily high opportunity cost.

Anomaly’s Stance: Artistic Intuition Over Programmed Logic

Our position prioritizes Human Agency while recognizing that Artificial Intelligence represents a massive opportunity to bridge gaps in society.

Instead of leveraging AI to amplify our own outputs, the trend to replace thinking is rising, perhaps led by the proliferation of Generative AI tools which prioritise quantity and speed. This leaves a vacuum for tools that prioritise human thought and voice, and a design space to explore for retaining creative control; technology which elevates, rather than erodes human intellect.

Introducing Paradox: A personal interconnected workspace to bring AI to your work.

Paradox is a product spun out from our own internal use. A personal workspace which includes an interconnected modern writing app, native AI chat and a generative canvas. It was created to support the way we develop and refine ideas: brain-dumping and planning visually, brainstorming with AI, aggregating references, suggestions and knowledge, and organizing thoughts before moving to writing and prototyping.

This process previously required five separate apps and ensured a lack of momentum, which Paradox solves by enabling users to start anywhere and move their work to any section of the app. Paradox prioritizes human voice and control while leveraging AI's strengths for repetitive tasks and rapid execution.

Knowledge also compounds as human understanding develops and branches over time. Instead of taking thoughts to a chat interface, we needed a system that would take the things we do daily, and compound them in an additive, controllable, reusable substrate.

This is where Paradox lies: a personal, living, connected knowledge layer that grows with everyday use. Everything users do results in semantically enriched, machine-readable data.

A connected workspace to accommodate any workflow:

Paradox Editor: A writing app built for the AI-age, it features a split-screen for human-AI collaboration without surrendering control to AI-generated content. Also included is a built-in Version Spine which enables exploration, and functions as a time machine for the continuous and branching nature of writing.
Native Chat: Talk to AI with your documents, previous chats, and topic-based context connected. Use bits or all of your conversation in your docs and canvas generations. Users can also import, highlight and distill information from previous chats with ChatGPT and Claude, and inject it into new chats, or new or existing documents thereby transforming segments of previous chats into enduring knowledge.
Paradox Canvas: A visual planning, control and generative surface, users can transform their writing into social media content, visuals, twitter threads, LinkedIn posts, blog posts and more instantly.
Multi-export: With native support for PDFs, JSON, Markdown, DOCX, HTML and TXT files, and integrations for Dropbox, Google Drive and Ghost, Paradox is a single platform covering most modern writing use cases.

LLMs can be high-value, cost-effective partners for facilitating human vision. When combined with Paradox’s connected workflow and compounding knowledge layer, they form a system where productivity is amplified and knowledge grows organically, while direction and control remains distinctly human.

Agents of Chaos: The AI Hyperbole Risk

Wed, 21 May 2025 00:00:00 +0000

No one can agree on what an AI agent is, but everyone’s selling one.

Buzzwords often do more harm than good and Agent is the latest in this list. As the article linked below highlights, the people and companies who create and fund Artificial Intelligence are unable to form consensus on the definition of Agent. How is it then that consultants, integrators and software companies are able to package and sell Agents?

TECHCRUNCH

No One Knows What The Hell An AI Agent Is

The tech industry continues to struggle with defining AI agents, as companies race to release products with contradictory definitions and capabilities under the same label.

March 14, 2025 Read more

Most Agents aren’t much more than workflow orchestration tools with an interface bolted on, and marketing them as such is a disservice to the industry today and in the future.

Anthropic, makers of Claude.ai, provide a good frame of reference to define agents

Agents have more autonomy which makes them inherently less predictable. While this freedom is wonderful for scope-defined research or calculating routes based on changing parameters in a fixed simulation, it is less than ideal for automation, reproducibility, customer-facing interfaces or where adherence to vision or values is important.

So, why the push for agents?

Table Stakes: AI Everywhere.

ChatGPT, Claude, Grok, Gemini, Meta AI, Mistral, Qwen, DeepSeek, everyone has an AI. Some are paid, others are free. Most are proprietary while the models originating from China are mostly free and/or open source which is refreshing to see as it democratises access and unlocks deeper innovation. The implication is commodification with little monetary value accrual.

Generative AI: Maturing, Multifaceted Technology.

AI trained on a large corpus of data is capable of mesmerising outputs. We marvel because generative AI bridges skills gaps immediately and is affordable for most people, historically two massive barriers to entry.

Large Language Models (LLM’s) have changed much of the writing one encounters. People use AI to write everything from essays to cover letters. You can use Suno.com to create music, Midjourney to create images, Higgsfield.AI to transform created images into jaw-dropping video. These are tremendous technological accomplishments and a massive neutraliser for people who didn’t have the training or opportunities to develop and refine their own skills. The fidelity and quality of these tools will only continue to improve and result in stunning work at a fraction of the cost when harnessed correctly.

Technology is not without its risk vectors though. When people use LLM’s to replace their thinking instead of improving it, mundane artefacts emerge. When music is uploaded for the sake of capturing streaming dollars, it dulls collective standards.

DEEZER NEWSROOM

Deezer Reveals 18% of All New Music Uploaded to Streaming is Fully AI-Generated

Recent analysis shows a significant increase in AI-generated music content across streaming platforms, raising questions about the future of music creation and artist compensation.

April 2025 Read more

Art - writing included - is creative energy shaped by emotion, philosophy and lived experience guided by intention. While some AI-generated outputs meticulously crafted by user inputs are truly excellent, most users are not incentivised by qualitative outcomes; their output which unfortunately constitutes a significant portion of the content available online, lacks the depth, coherence and intentionality of artistic expression.

This compromised artistic integrity creates standard deviations between intention and outcome which are a form of emergent chaos, more damaging than noise as they risk stagnation and the standardisation of what will eventually become mediocrity.

The Rise of the “Agent” Label

The Agent label today falls into two categories.

Guided Execution:

Scoped applications of technology where AI equipped with tools works semi-autonomously to produce outputs, examples of this are Deep Research and instances of agent-assisted coding like Codex, Claude Code and Cursor. The quality and capacity of the latter is debatable as the technology is still in an early stage of development and like generative AI, will continue to improve and should be harnessed to augment human effort and amplify productivity.

Sensationalism:

Lack of Authenticity: As technology commodifies at the base layers, it loses its unique value proposition. Similar to email providers, most users find AI tools interchangeable as they all serve the same basic functions. Entities with no authentic value to offer or layer on resort to sensationalism to attract attention and generate interest.
Redundancy: Free Generative AI can provide strategy, visual and written content, planning and ideas which were previously gated by access and/or knowledge. Organizations and individuals who stand to lose revenue are incentivised to maximise profit while they can.

Sensationalism is a symptom of mediocrity. In this sense the term Agent is akin to ostentatious thumbnails and clickbait titles. An over-the-top thumbnail doesn’t lead to high-quality content any more than a social post designed to go viral enriches its audience. Worse, it attracts opportunists and grifters both of whom add noise and obfuscate signal. Entities selling agents stand to gain from doing so by capitalising on industry-wide momentum, or wanting to differentiate their undifferentiated and often unnecessary offerings.

As emerging technology changes the landscape, new solutions must emerge and take centre stage and legacy systems must recede, otherwise stagnation ensues. Not their algorithms, we need new surfacing mechanisms which don’t prioritise the provider; ones which scan for signal while filtering for incompatibility and undesirability.

Anomaly’s Stance: What True Agency Requires

Our stance is that AI Agents are a post-personalisation paradigm. True Agency in the context of Artificial Intelligence requires:

Trust: a less abstract representation of which is predictability. Today’s outputs are unreliable, at times undesirable and unpredictable. If an undesirable output is predictable, then the present solution represents design failure or incompetence.

Personalisation: Understanding goals, values and context. Generic models cannot serve as effective agents because they lack a persistent contextual understanding, which is a combination of long-term memory, deep contextual awareness and adaptive learning, amongst others.

Security and Safety: Agent contributions must be structured to be additive, safe and private. Current architecture and prevalent business models are conceptually incompatible with this ethos.

Developing these capabilities is currently costly, resource-intensive, and difficult to scale . Training or fine-tuning models per user is computationally expensive. Maintaining stateful interactions and memory adds complexity and most companies aren’t ready to invest in the infrastructure needed for real agent-like systems when the technology arc hasn’t reached maturity.
Even in the B2B context, it is a reenactment of the Cloud-based vs on-prem debate as organisations require personalisation - knowing their structures, workflows, repositories - for the technology to be truly effective, and this has security and dependency implications.

Trustworthy AI Agents will require true personalisation and a rethink of monetization and ownership models.

The Real Costs of the Hype

The rush to label everything as an "AI Agent" risks:

Lowering expectations for what real agency should be.
Misleading Investors and wasting funds.
Setting negative precedents and systemic obstacles for aspiring entrepreneurs in the future
Encouraging dependency on systems that lack accountability.
Distracting from the harder, more important work of building genuinely helpful, consistent, and ethical AI assistants.
Reconstructing old paradigms. Chatbots risk recreating the IVR/Call Centre loop of frustration.

The Road Ahead:

Agents will ideally coexist as collaborators, personalized in understanding, tempered in ability.  For this vision to come to fruition, we must channel our energy towards:

Data security and context engineering
Memory architecture and methods
Protocols that transcend privacy limitations
Architectures that support persistence and emergent capabilities

At Anomaly, we believe that personalization is the key unlock for Artificial Intelligence and that this is the fulcrum where the industry will ultimately converge. This belief forms the foundation of our thesis, and it guides our work in long-term memory, context-aware models, user-specific tuning methodologies, and ethical frameworks, so that when agents do emerge, they enable humans to build a safer, empowered and more equitable future.

AI is a broadly available infrastructure that is evolving and permeating. Given its magnitude and the role it will play in tomorrow's systems, its application must be ethical, resilient, performant, and focused on ingenuity, as new paradigms mandate expanding our perspectives to identify, accommodate, and address increasing opportunities and vulnerabilities.

Trivial Pursuit? - Jobs In The Age Of AI

Sun, 18 May 2025 00:00:00 +0000

Much has been said about AI displacing jobs, much continues to be said. Given the magnitude of the discussion, proper framing is warranted.

Technology is a tool, sometimes borne of necessity but mostly of the human capacity to imagine, which puts more within human reach and has the power to uplift or tear down, to build or destroy. AI is a multimodal tool with reasoning abilities and unprecedented accessibility. It's available everywhere and mostly subsidized to increase engagement. Meanwhile, a job is the regular work that a person does to earn money.

A quick glance at history verifies that technology displaces jobs but creates many more as it runs its arc. It spawns entire new industries which branch out into others.

A machine can work cheaper, better, faster and longer than a human can, about this there should be no debate, but this does not mean we teleport to Universal Basic Income because everyone will be out of work.

What is Work?

Let's try to answer the question: Why do we work?

This diagram attempts to answer the Why. The reasons are many, and while some matter more than others, we do it to make ends meet, secure our future, be a part of something and many others loosely related to Maslow's hierarchy of needs.

What about the flip side? Why do companies hire people (Jobs)? They do it to fulfil a specific set of requirements (purpose), and are incentivized to maximize profits. i.e for the lowest cost.

So jobs are created to fulfil a purpose, and this is usually accomplished by humans executing specific tasks. Viewed through the lens of systems, these reasons combined represent the System Level (purpose) which extends to the Execution Level (Job) and the Method Level(Tasks). These map to the Why, the What and the How.

Does Work Stop?

From post dotcom billionaires interviews to pre-iPhone studies, there is plenty to substantiate the claim that people work, and continue to work long hours, when they don't need to.

NBER DIGEST

Why High Earners Work Longer Hours

Research reveals a dramatic reversal in work patterns: by 2002, the highest-paid 20% of workers were twice as likely to work long hours as the lowest-paid 20%, driven by increased "marginal incentives" for extra hours worked beyond the standard 40-hour week.

July 2006 Read more

Now, as a current-day thought experiment, imagine everyone is economically secure, socially content, adequately credentialed and has an abundance of time. Let's even add Universal Basic Income in the mix.

Do humans stop working? Do writers, actors, engineers, bakers, photographers, thinkers, entertainers all call it quits? No, the very notion is absurd.

We work because we possess creative energy which requires expression, and curiosity that demands action; the power to imagine and the desire to pursue that vision’s fulfillment.

Creative Energy is the fundamental substrate other levels serve to channel and direct. Systems are energy-organizing structures which channel energy as a conduit to create value. Across industries, premiums are commanded by the most creative - those who have a way - whether they are lawyers, chefs, consultants, doctors, tennis players, musicians or anything else. The demand is endless, and the domain - creativity - is uniquely human.

Creative problem-solving and strategy - a skill that requires creativity and context - are not the direct domains of AI which can only mimic creativity through pattern recombination. Large Language Models (LLMs) are just that; language models, and before compute is intensified to implement problem-solving, the domain that is easiest to disrupt is language. The real-world implication of this encapsulates computer languages and written language. LLM's will write code and words more efficiently than human beings which means industries that rely on those skills will be impacted first, but not where creativity is required.

Having the ability to harness an LLM to code faster and correctly does replace software developers because the incentives align for businesses and the effort to do so is low, but it also produces many more software creators who previously had an execution limitation. In other words, automation reduces the cost of the function, but increases the demand of the output while creating new, higher-level jobs that were not possible before.

As such, AI is unlikely to spawn novel creative digital products and services, while humans will certainly leverage AI to do so, and in turn, inspire and hire many other humans.

Facilitation i.e. low creative energy work - organizing existing information, following established processes, coordinating routine tasks - maps well onto what current AI systems do effectively. These are scoped areas where the intelligence required is more about pattern recognition, repetition, rule-following, and optimization within defined parameters.

This is a long-winded way to say that AI will not displace jobs, rather displace stagnated energy. A suboptimal job is essentially inefficient energy allocation; human creative capacity trapped in work that doesn't warrant it. Displacement in these cases need not be destructive, but can be liberating as it forces the energy to find better outlets.

The internet gave users distribution thereby liberating siloed information, Generative AI is now bringing sophisticated creative capabilities within reach. Coupled with creativity and strategy, this will likely spark many new creative projects and expand the creative class as an aggregate percentage of new enterprises[system-level organizing structures] serving an expanding marketplace.

Anomaly’s Stance: Human Energy Architecture

We believe Generative AI is an equalizer and will create more artists and entrepreneurs.

Yes, there will be more noise to sift through but AI is a net positive for Human Energy and Agency.

No, we are not doomed. Machines cannot think and genuine creativity is the exclusive domain of humans. Instead of just reskilling for the AI-era, we have the opportunity to optimize human energy allocation by helping people discover latent potential.

We anticipate a rise in entrepreneurship and solo ventures as AI enables those with vision, perseverance and fortitude to forge ahead. We also expect some traditional employment to oscillate from single and static, to fractional and fluid. This paradigm will likely be very different from gig work in that it'll be based on individual contributor strengths rather than interchangeable contributor abilities.

We expect the future to spawn many new employee-centric methodologies and ecosystems as we navigate to a world where people’s unique energies find expression in work that matters to them.