Malleable Software Reshapes the World
How AI Alters Creation, Enterprise Structure and Global Markets
Introduction
As highlighted by Geoffrey Litt, after nearly 40 years, computer pioneer Alan Kay's vision of end-users harnessing the full power of computing, not just using pre-packaged apps from a programmer elite, is finally coming to fruition. Generative AI tools and AI agents are dramatically lowering the barriers to software creation, enabling everyday users to tailor and generate software that perfectly matches their needs. In doing so, we are entering a new era of software, one in which users themselves can easily shape, extend, and customize, not just at a surface level with pre-built toggles but at a complete and fundamental one. We enter the age of malleable software.
Powered by generative AI and often executed by AI agents, malleable software enables all users, including experts and non-experts alike, to create, modify, and extend digital tools in real-time, often simply by providing a natural-language prompt. This shift will blur the lines between "user" and "developer," fueling an unprecedented wave of creativity, rapid iteration, and business transformation. It is more than just an extension of existing trends but a step change that will rapidly alter how we think about products, work, the nature of the firm, and market structures.
Over the next few years, we will come to rethink the nature of production, the shape and form of the enterprise, and see a massive reordering of capital markets activity to support this extraordinary change.
Part I: Understanding Malleable Software
From Fixed Code to Adaptable Products
Historically software has mirrored most products and followed a well-defined production cycle with defined barriers between those who create it (the developers and the companies they form) and those who utilize it (the users): developers write code, testers check it, and companies ship a final product to users who then operate the product primarily as provided.
Outside of minor changes such as interface color themes, end users have had minimal power to alter an application's functionality. Only those with specialized coding skills could build or significantly modify digital tools. In fact, over time, as code complexity has increased, the time and cost for engineers to effect changes in end products have also increased. Today, very few large scale products are made by solo enterprising individuals. Moreover, even those with technical expertise and time have been limited by the exposed endpoints of a given product when it comes to customizing the product, requiring significant investment of time and resources, thereby limiting experimentation.
AI overturns these constraints. Through large language models (LLMs), natural-language prompts, agents, and AI-driven app architectures, it becomes possible for end users—often with no formal coding background—to shape software on the fly. Today's containerized, microservice-based, and extensive API-supported enterprise and consumer applications serve as merely a façade of software interactability. But by combining these structures with AI "co-pilots" and capable semi or even autonomous agents, generative models turn software into a collection of building blocks that can be rearranged or even rearrange themselves at will. Products become fluid, software becomes malleable, and applications/UIs evolve continuously to meet user needs.
Imagine a casual user who opens their favorite music app and starts by telling an AI assistant: "I want a minimalistic dark theme for this app." Instantly, the interface adopts a sleek, dark-toned design—no manual coding or fiddling with settings required. Next, the user says, "Add dynamic menus so I can quickly browse between playlists, friends' recommendations, and trending tracks." The app instantly rearranges its navigation elements, rolling out collapsible side panels or quick-swipe gesture menus in real time.
Seeing how easy it is to customize, the user proceeds: "Whenever I play a song, please pull real-time lyrics and display them under the track." In seconds, the AI integrates the necessary API calls, turning the once-static Now Playing screen into a live lyrics feed. The user then decides they want to sing along and asks for the interface to shift into a Karaoke experience complete with judging and feedback on their performance. Not quite satisfied with that, the user then requests to shift the octave and speed up the music, remixing it into a different musical genre. Finally, the user decides to link this experience with external data: "Connect my calendar so I automatically see local concerts for the artists I've listened to this month." The AI updates the code again, taps into event listings, and creates an in-app widget that highlights upcoming performances. "Now monitor for any great deals and purchase tickets for me and my fiancé to any concert that fits our schedule and budget." A persistent agent is spun up to manage the monitoring and purchase of tickets in the user's sweet spot.
At every stage, the user speaks or types their intent, and the application adapts on the spot—no specialized development team or months-long release cycle needed. Moreover, at the end of the process, an autonomous agent is spun up to extend the user's preferences beyond the present, adapting on the fly to changes in circumstances and new data. This step-by-step transformation illustrates the power of malleable software: a world where anyone can meaningfully change both the look and core functionality of their software, tailoring it to fit their individual preferences or even create new software to fulfill their goals. In this reality, product managers and application developers no longer need to predict exactly what end users will want, but instead create the space for users to ask for what they need.
We are rapidly nearing a point where everyone will be able to create and remix software.
Part II: The Consumer Landscape: User-Driven Mods and Endless Variety
The “Modder Economy”
While the growth of malleable software reflects a step change in the ability of end users to alter and adapt software to their needs, it sits firmly within the rising tide of increasing user-generated content.
Our expectations of UGC continue to evolve. It's evolved from simple photo sharing to elaborate video platforms and user-created gaming experiences. With each wave, think early social media photos, then YouTube and TikTok videos, and later entire game worlds, barriers to creation have lowered, allowing everyday people to generate vast amounts of content of increasing complexity and sophistication. Moreover, users have been trained to want to remix and contribute.
We have ample examples that show people, when given the ability to create, do so. Whether it's billions of daily photo uploads or hundreds of hours of video posted every minute, UGC platforms thrive on the sheer volume of user creativity. In 2023, users posted over 10 billion videos on TikTok, and that figure is projected to have more than doubled in 2024. An astounding 52% of U.S. accounts have uploaded a video, and overall, TikTok supports hundreds of millions of creators who actively shape trends and content. Snapchat similarly opened its augmented reality toolkit to the community, and over 300,000 community creators have produced more than 3 million "AR Lenses" despite limited economic incentive.
Arguably, the most striking example of user-driven creation of complex software products is Roblox. The platform—known for its "create anything" ethos—hosts over 66 million daily active users, many of whom are kids and teens building and sharing entire game worlds. In 2023, U.S. creators on Roblox collectively earned over $740 million from content they built and that number grew by 25% to $923 million. Notably, 44% of these Roblox creators stated that they would not have created interactive experiences at all if not for Roblox's accessible tools. As such, we already have numerous examples of users creating increasingly sophisticated products using rudimentary tools that fall short of the power of modern AI-based creative tools.
Now, malleable software builds on that foundation by granting users the power to alter not just visual or audio elements, but the core functionality of all software itself, and to do so not just within the confines of walled-garden social platforms, but across every piece of software.
With generative AI, user creation will no longer be limited to recording a video or changing a game skin—people will customize entire workflows, app features, and cross-platform integrations on the fly. Rather than dance videos or kids' games, user-driven creation will extend to every corner of software. A local business owner might integrate a custom reservation system into their social media, while a fitness enthusiast might embed personalized workout stats into their music streaming service. Malleable software removes the final barrier—coding expertise —and frees users from the confines of their walled-garden platforms, thereby unleashing the full creative potential of everyday users.
Everyone will become a modder.
Of course, some businesses may initially resist such changes. Even before the rise of AI coding and browser-capable agents, more sophisticated users were able to modify software, build scrapers, and other tools to, in effect, turn websites and other products into “APIs.” Businesses have sought to block some of these applications, e.g., ticket scalpers with CAPTCHAs and custom game mods with cease-and-desist notices. These measures have been successful to varying degrees. And when it comes to AI we are seeing the same pushback from walled gardens, see Slack blocking others from searching and storing messages, even if their own customers grant them permission.
However, as the tools to build malleable software proliferate and as we increasingly rely on AI agents to perform tasks for us, the equilibrium will shift, and developers will be forced to open up their products to agents and malleable software modders. In a world where modification and agent-based access are the expectation, developers who do not open their products up will be at a severe disadvantage compared to those who do. Moreover, as the tools become more sophisticated, it will be harder to harden applications and products against modification and agent access. Eventually, most developers will aid the rise of malleable software and products that enable greater modification and will outcompete those that do not. Already, we are seeing a shift towards an AI-friendly web with the rise of MCP servers and the push to build an "agentic web" from the ground up, encouraging AI-based agents and other software to directly access data and services in a machine-readable manner that supports malleable and customizable workflows.
What Even is a Product?
Equilibrium Shifts: Redefining Product in a Malleable Software Era
As software becomes increasingly malleable, user-driven modification commonplace and everyday people create software agents to extend their abilities, the very concept of what constitutes a product will inevitably shift.
Until today, most software applications have been tightly bundled combinations of user interfaces (UIs), databases, and programs designed to manipulate these databases. Each component was intricately linked, with changes requiring significant effort from specialized developers. Malleable software breaks apart these traditional bundles
Initially, we will see the UI become significantly unbundled leading to the “APIification of everything.” Traditional interfaces will no longer be the sole, or even primary, way users interact with software. Instead, interactions will increasingly occur through sophisticated AI agents and adaptable interfaces customized by end-users. Users will guide software via natural language, voice commands, and contextual prompts, bypassing rigid and pre-defined navigation pathways. This transition represents a fundamental shift in product design philosophy, emphasizing flexibility, user intention, and dynamic adaptability over fixed graphical layouts.
Next, the underlying software programs themselves will start to become more modular and unbundled. Instead of relying on pre-defined features and fixed functionalities, users will begin to effectively “write” or "compose" their own software components and then entire agents, assembling personalized tools and workflows. Powered by AI-driven tools, these user-composed programs will interact fluidly with existing databases and back-end systems, further dissolving traditional barriers between developer and user.
In this evolving landscape, the concept of a software product will shift towards something akin to a collection of agents, data feeds and standardized protocols rather than monolithic, fully packaged applications. Products ought to and will increasingly serve as flexible infrastructures upon which users build personalized experiences and functionalities. In many cases, users may not even directly or even knowingly interact with these products, instead choosing to have their agents do so instead.
This transition won’t eliminate the importance of great design or intuitive user interfaces—on the contrary, the most successful products will empower users and their agents to easily and effectively personalize their software experiences.
Not all users will desire or need to actively modify their software. Many will continue to prefer products that provide exceptional default experiences. Thus, the key to success in the malleable software era will lie in balancing powerful default designs with robust, user and agent-friendly customization and integration capabilities. Software creators must embrace the dual challenge of designing both universally appealing default experiences and seamlessly enabling personalized adaptations.
Taste vs. Technique
As Benn Stancil suggests, as software becomes malleable and software production becomes democratized, we will see a fundamental shift in the nature of who creates software and who is rewarded for their efforts.
Thanks to AI, "taste"—knowing what to build, increasingly matters more than raw coding technique or skill. Malleable software and AI more generally, has the effect of increasing the raw supply of “technique” or more simply put capability. As the supply of technique goes up, the returns flow to its compliment: taste.
If the previous era rewarded, deep programming knowledge and a team of engineers to bring an idea to life, the next era, with AI handling much of the heavy lifting, will reward someone who only dabbles in code as they can spin up a functional prototype in days—sometimes hours—merely by describing what they want to build. If the software can be created so quickly, the real bottleneck is not coding skill—it is the underlying insight, creativity, or domain familiarity of the person designing the tool.
We've seen parallels in virtually every major user-generated content arena: the biggest YouTubers aren't typically Hollywood directors, and the most popular TikTok creators don't come from traditional film schools. Likewise, the top Roblox developers are often young enthusiasts who grew up on the platform, rather than veteran game designers from legacy studios. When these amateur creators rise, they usually do so because they have a better pulse on what resonates with everyday users, not because they hold impressive credentials or come from elite backgrounds and formal training. Simply put, the skills needed to succeed in a more democratized field are inherently different. The skill required is taste.
Taste is an exciting sea change. Instead of living in a world where output was constrained by who had the skills, funding, and connections to bring an idea to life, once creation tools open up, we discover that people from varied backgrounds—often without formal training—can produce experiences that others prefer not despite of their differences from historical producers but because of them. Their unique insights, experiences, and intuition can trump the carefully crafted assumptions of "experts."
This dynamic shifts the playing field for product creation. If passionate individuals with an insight into architecture, entertainment, healthcare, finance, or virtually any domain can quickly translate their "taste" into working software, the advantage that once belonged solely to well-trained engineers begins to wane. It's no longer enough to be able to code; you must also have the vision, nuance, and genuine understanding of a problem space to craft something truly valuable. Crucially, those with no "elite" training but strong familiarity with a particular niche or audience can be just as successful, if not more so.
Of course, many "experts" will initially dismiss these grassroots creations as lowbrow or "slop." Already, there is pushback from these "elites" - they decry the inevitable rise of what they see as thoughtless, careless, and undifferentiated, lowest-common-denominator products. Yet, the market still frequently shows that what resonates with the masses may differ sharply from what a small circle of professionals deems high-quality. While it is true that many creations powered by the democratization of production will indeed be of low value, with minimal effort or care, this perspective ironically overlooks the forest for the trees.
History is littered with experts and elites decrying the production of the masses, see Hollywood's reaction to YouTube or even the reaction to printed theology from the Catholic church. However, elites' "slop" is often the masses' treasure, and the value defined by elites is often just credentials and distributional advantages, masquerading as product advantages. Even if those elite values reflect real advantages or “quality differences,” consumer appeal is often different than producer appeal and consumer taste" and "producer taste" are actually different. If you spend years developing technique, it impacts who you are and how you think about quality. You understand difficulty in a much more nuanced way, and so "producer taste" tends to be much more technique-oriented. In some sense, producers tend to have more similar taste to one another than they do with the median consumer.
The point of democratization is not that every product produced by the crowd surpasses those bestowed upon us by the elites but rather that inevitably some will, at least when it comes to appealing to a broader audience less focused on “peacocking” and the elegance of a solution and more focused on fundamental needs.
More broadly, even in defining these products as low effort or created with minimal care, we are mistaking the old paradigm for the new, in which the hard work is not done at the point of generating the code, but in gaining the experience and knowledge to decide to develop it in the first place. Why is an insight gleaned from years of real-world experience, a design idea driven by intrinsic instinctual knowledge honed by years of practice, lower effort, or less valuable than one built via a long, painstaking HBS-approved process? The power of AI is in shifting the effort away from bricklaying to dreaming.
Regardless of elite opinion, many of the ideas that used to remain "what-ifs"—because non-technical people couldn't build them—will now come to light. Some will fail, as many ideas do. Still, others will flourish in unexpected ways, precisely because they originate from real-world experience rather than abstract brainstorming and unique insight, as opposed to expert groupthink. Ultimately, in an era where coding overhead continues to shrink, taste and perspective will become the real differentiators. Products that reflect a genuine understanding of user needs will outshine those generated by purely technical skills. The future of malleable software belongs to those who bring both creativity and substance, not just process and credentials.
Product Distribution, Security, and IP in a Malleable World
If software becomes increasingly malleable, and taste ultimately outweighs sheer technical skill, what new challenges emerge in a world of software abundance and lack of software scarcity?
Distribution Will Matter More Than Ever
In a world where creation is abundant and cheap, beyond "taste", a significant differentiator becomes how effectively creators can reach audiences or customers.
As coding skill ceases to be the main barrier, those with access to established communities, proven marketing channels, or unique distribution partnerships will hold an edge. While creation will be abundant and easy, many users still find a blank canvas overwhelming. Products and people that solve the blank page problem or direct users to the best products will still win.
We will likely see new, specialized "App Stores" or distribution platforms for malleable software and AI agents, where creators share their customized experiences and most valuable agents. Even these platforms themselves might be malleable, allowing users to remix storefronts, rating systems, or recommendation algorithms in real-time.
Social proof, community curation, and discovery will continue to be essential for guiding people toward popular or high-quality content, while also exposing them to niche ideas. Because so many users crave examples and reassurance about what works, these platforms will help them find, adapt, and refine software mods and agents that others have already validated. Without these social signals, users might be overwhelmed by an endless sea of software, unsure where to start or which version to trust.
As such, we can expect an even greater rise in the importance and reach of influencers across all domains who leverage these new malleable capabilities to create, remix, and distribute at scale. Their strong connection to audiences—whether mainstream or hyper-niche—will let them stand out in a sea of infinite content. This dynamic will further hollow out the middle: specific creators and brands will achieve massive scale thanks to the ability to experiment, build, and distribute increasingly sophisticated products at minimal cost, while niche experts will develop even more loyal, focused communities around serving highly specialized needs or desires. Those who produce merely "okay" offerings in an oversaturated market will get further squeezed out.
More radically, builders who can develop strong distribution pipelines that "appeal" to the AI agents operating on behalf of users will similarly see outsized success. While we don't yet have a firm grasp on what building for agents looks like exactly, it is clear that doing so well will confer a huge advantage. Already, we are seeing the rise of Generative Engine Optimization (GEO), which involves optimizing output to appeal to AI agents through improved schema, data, and content structure. This practice will only become more intense as more users create more agents to operate on their behalf, and more global search and software usage becomes agentic. Regardless of its form, however, the returns to differentiated distribution will likely only grow.
Security Grows Ever More Complex
Over time, as demand grows for malleable experiences and agent-optimized applications, product owners will need to pull down their walled product gardens to compete.
Allowing end users and their agents to modify code, integrate external data, and reconfigure core functions inevitably raises security concerns: how do you protect user data or brand identity when practically anyone can alter an app's logic on the fly?
Striking a balance between encouraging creativity and preventing malicious tampering will become harder. Software vendors may lose some measure of "control" over their products, as user-run AI agents orchestrate new workflows or integrations behind the scenes. Platforms will need robust permission systems, real-time scanning for threats, and an adaptive trust model that can handle a fluid ecosystem of add-ons and remixes—none of which may be officially vetted. We will likely need to rethink our notions of security and identity. In many cases, enhanced security and greater trust and safety may be key product differentiators that enable more specialized businesses to outcompete more generalized AI-based solutions.
The Changing Nature of IP
When users can freely remix and modify code, visuals, or even entire services, questions about intellectual property become more thorny. We've seen parallel debates with user-generated music remixes and fan-made game mods: where does homage or sampling end, and infringement begin?
As malleable software grows, this tension intensifies. Traditional licenses may not easily accommodate scenarios where "original" code or designs are continuously built upon, rebranded, or mashed up with features from multiple sources. Moreover, value could flow more to the best remixer or integrator than to the initial IP owner. While IP will still matter, controlling it becomes more difficult, and successful models may revolve around open collaboration, revenue-sharing frameworks, or hybrid licensing that encourages remix culture while preserving core rights. Already, we are seeing large organizations attempt to quash new AI-written software "impeding on their IP". While this may work for now, the history of UGC content suggests that this strategy will be about as successful as sticking a finger in an increasingly leaky dam. We will need new legal structures and monetization schemes that embrace, rather than fight, AI-powered remixing.
Part III: Enterprise and Organizational Transformation
The rise of malleable software and AI agents will not only shift consumer markets and individual behavior but also dramatically reshape how enterprise organizations function.
Traditional economic theory views firms primarily as entities designed to reduce transaction costs by internalizing coordination, resource allocation, and decision-making. However, as software becomes increasingly flexible, easily written, adapted, and customized by broader groups of users, it profoundly changes these assumptions. Moreover, as users become capable of creating their own autonomous or semi-autonomous agents, the very nature of employment and work shifts, and everyone becomes a manager or orchestrator. Coordination costs fall sharply, specialization becomes less rigid, and the necessity for centralized human management diminishes, fostering the emergence of flatter organizational structures, agile teams, and fluid, on-demand workflows. Malleable software and AI agents will therefore give rise to very different types of enterprise businesses and profoundly reorder market structures.
Companies as Transformative Programs
At their core, companies can be viewed as dynamic systems composed of programs—processes that orchestrate the transformation of less valuable inputs into more valuable outputs, utilizing resources such as dollars (money), watts (energy), atoms (material), and bits (information). Historically, these organizational "programs" required significant human intervention, coordination, and oversight:
Primitive Programs: Entirely human-driven, relying on biological energy and manual coordination. Often individual and highly limited in both structure and output.
Agricultural Programs: Energy from domesticated animals transforms plant atoms, allowing and requiring more complex organizational forms.
Industrial Programs: Mechanization replaces manual and animal labor but still requires extensive human oversight.
Software programs: Code begins to replace some elements of human supervision and intelligence by encoding information in machine-readable bits. However, ultimately, these programs are still built, modified, run, and overseen entirely by humans. Most critically, only certain activities and processes, primarily those capable of being encoded in machine-readable language, can be transformed into software programs. Moreover, most software programs require human intervention when interfacing with the non-digital world or unstructured digital data.
Each stage required substantial human effort in planning, execution, and adaptation. But we are on the cusp of a new era. AI excels at transforming unstructured human activities—such as informal communications, creative workflows, or customer interactions—into structured, machine-readable data. As such, AI systems can analyze this data, identify patterns, and translate these activities into automated workflows executed by AI agents. This means activities previously considered too complex, nuanced, or human-dependent to automate can now be increasingly encoded into software, and then executed by software, dramatically expanding what can be automated and optimized.
With AI agents, tasks traditionally handled manually or semi-manually—like scheduling hospital staff, managing airline logistics, or orchestrating manufacturing processes—become fully digitized and automated. Rather than merely observing business operations through static dashboards or analytics, operational logic itself is embedded directly within AI-driven systems and agents, automating decisions at machine speed and dramatically shortening the distance from insight to action. Moreover, by enabling machines to execute actions that previously relied on unencoded insights, data, or decision-making, AI creates significantly more leverage for remaining human activity and management.
Already, agents and AI systems are capable of analyzing and executing a wide variety of tasks and processes, primarily digital and requiring relatively few subtasks and chains of action. However, as agents become more sophisticated and fundamental models become more powerful and as multi-agent workflows become more accessible, the complexity and length of tasks that can be machine-encoded and executed will grow exponentially. Moreover, as these systems grow and more infrastructure is created to support them, particularly controllable robotics, they will be able to reach out and encode and automate real-world tasks in addition to purely digital ones.
Ironically, as real-world industry is transformed and made more efficient through AI-powered software, software will revert to more industrial cost scaling, given the marginal cost of inference, further blending the digital and physical business worlds.
Malleable Software as an Engine of Business Democratization
AI, therefore, dramatically increases the scope of the types of programs we can build and automate as software. At the same time, AI agents themselves create the means to massively increase the leverage and efficiency of these programs both digitally and in the real world. But just as critically, as software becomes malleable and these tools reduce the friction involved in modifying and optimizing these business programs, malleable software opens up that capability to a much larger group of people at a fraction of the total operational headcount and cost.
Rigid workflows, currently complex to develop and costly to alter, become easily accessible and editable by diverse stakeholders, significantly democratizing the capacity to innovate. Rather than relying solely on internal or external developers with lengthy development cycles, organizational workflows can be quickly assembled, adjusted, or repurposed by employees at various levels, fostering widespread experimentation and rapid iteration. Simply put, every employee becomes a forward-deployed engineer. Already, startups such as Workflow86, a YC company, are providing non-technical users with the ability to merely describe processes in natural language to create AI agents to automate their tasks.
Because companies fundamentally consist of these programs and processes, the increased malleability of software and the further democratization of software generation and management will directly translate into enhanced organizational efficiency, as well as greater flattening and fluidity. Successful organizations will rapidly adapt their workflows to meet market demands, regulatory changes, or internal feedback, without significant downtime or investment.
Or they will be outcompeted by those who do.
Legacy companies will either adapt or die, and overall enterprises will become not merely agile, but inherently dynamic—able to continually reshape themselves according to evolving circumstances and do so more efficiently than ever before.
Rethinking Organizational Boundaries
Traditional economic theory asserts that firms internalize processes to minimize external transaction costs.
Given that malleable software and AI more broadly significantly reduce coordination friction and massively reduce costs while expanding the scope of information transfer and transformation, we assume it will prompt organizations to not only reimagine their internal processes and structures but also to reassess traditional external boundaries. Internal and external capabilities become fluidly interchangeable, enabled by seamless API like integration and real-time software adjustments, managed by and executed in part if not wholly by AI agents. Organizations will dynamically shift between insourcing and outsourcing based on immediate operational efficiencies, further blurring distinctions between internal and external processes. We expect the shape and outline of firms to shift dramatically, with some shrinking rapidly and others expanding significantly.
The Rise of the Cybernetic Enterprise
Together, these changes will transform previously rigid corporate hierarchies into responsive and adaptable ecosystems. Organizations leveraging malleable software will continuously experiment with new operational workflows while managing fleets of AI agents to execute them. Decision-making will transition from rigid, top-down pipelines to fluid, adaptive feedback loops collaboratively managed by employees and their machine agents on the front lines. Enterprises will expand and contract in near real-time as business fundamentals dictate. Businesses will be characterized by integrated, dynamically configurable operational processes more akin to APIs than org charts. Jeff Bezos's vision at Amazon of encoding every operational interface finally becomes reality.
This isn't simply akin to adopting new SaaS tools but rather represents a step change in organizational structure and capability. By making software inherently accessible, adaptable, and customizable by a broader range of organizational participants, malleable software effectively democratizes decision-making and innovation, not just at the consumer level, but also at the enterprise level. Companies become more open, participatory, and, most critically, dynamic, continually reshaped by the collaborative input of their workforce and AI-powered machine systems.
As such, with the advent of malleable software, we enter a new era, characterized by a new type of program: cybernetic programs.
In this era, human intelligence collaborates seamlessly with machine intelligence across the entire enterprise. This partnership fosters continuous feedback loops that dynamically refine operational processes and enable rapid responses to changing conditions. The enterprise nervous system is decentralized and pushed out to its constantly evolving edges. Organizations themselves become software, capable of being written and rewritten on the fly by the people and AI agents that represent them.
Part IV: Malleable Markets and Capital Flows
Suppose malleable software changes the nature of consumer software creation by letting anyone build or remix applications, and transforms enterprises themselves into adaptive, cybernetic organizations. In that case, it also stands to reshape broader market structures and capital flows.
Already, the cost of software-driven innovation has plummeted. Meanwhile, the line between "consumer" and "producer" has blurred, and the boundaries of firms are more fluid than ever. These changes will necessitate very different market structures and capital flows and require rethinking how and which projects, individuals, and companies are funded.
A More Adaptive “Main Street”
In traditional industrial-era economies, many real-world transactions remain slow or dependent on manual processes—think procurement cycles, supply chain negotiations, or regulatory compliance. But as more of these tasks become malleable, software-driven, and amenable to rapid reconfiguration, "Main Street" economic activity may begin to resemble the dynamism of capital markets, where resources flow more readily to where they are needed:
Faster Flows of Value: When software is inexpensive to build and adapt, new ideas can emerge instantly, pivot if needed, and scale or expand in terms of impact and revenue without the growth in overhead associated with large, specialized teams. Much like capital markets allocate funds to promising assets in near real time, malleable markets can quickly direct attention, users, and resources to the best ideas, products, or software experiences. We can expect to see a rapid decline in the time required to scale ideas, organizations, and markets. Already, we are seeing startups and AI-adopting organizations scale much more rapidly than ever before, thanks to the ability to push software production and automation to the front line of every employee. Agents and AI-based tools are significantly expanding the scope of what can be achieved and how quickly it can be accomplished. As such, we are seeing companies reach significant revenue (i.e., over $100 million) in record time and with relatively few employees. This is compressing funding timelines, increasing early-round sizes for successful enterprises and products, and leading to the rise of small teams and individuals achieving enterprise-level exits or outcomes. We should expect more rapid scaling and iteration across the board.
Reduced Middle Layers: In a world where building is easier than ever and side markets—i.e., creation, integration, and distribution—are more open and crowd-driven, the role of traditional middlemen or gatekeepers diminishes. Similar to how decentralized finance can match borrowers and lenders directly, malleable software ecosystems can connect creators with users, or employees with tasks, without as many organizational layers. As democratization increases across the board, we can expect to see a continued decline in intermediaries of all kinds. Capital will be directly connected to producers, and distribution with the product.
Ephemeral Organizations: The fractal, modifiable nature of these systems means we may see short-lived alliances, microservices, and ephemeral "pop-up" teams that form and dissolve as needed. Like traders who gather briefly around a new stock or commodity, these dynamic collectives can emerge in response to shifting consumer demand or changing market signals. We should expect enterprises and producers to operate more like locusts, swarming towards value, building and capturing opportunities in real time before either shutting down or moving onto the next project. We should expect enterprises and individuals to scale massively and quickly, only to effectively disappear. This should not be considered a failure but rather a fundamental part of the ecosystem.
Faster and Easier M&A/Divestiture: So much difficulty and cost in merging and divesting companies or business units is really "refactoring their programs" to use similar processes, procedures, systems of record. But as companies become more malleable and cybernetic much of this difficulty and cost should be reduced. You can even imagine software that specializes in "mergers and acquisitions" systems re-architecture and integration.
Toward a Liquid Innovation Market
Capital markets have long served as a signal mechanism for the economy, quickly pricing in new information. With malleable software, we edge closer to an environment where continuous creation and real-time feedback loops enable producers to respond to demand just as quickly, whether in consumer or enterprise markets. As such, we should expect some pretty radical changes to the structure of industry more broadly and a movement towards a more liquid and capital markets-like structure and operating model for the broader economic system:
Market Speed and Fads: Rapid iteration does not guarantee perfect outcomes. We should expect misallocations, hype cycles, and short-lived fads, just as capital markets experience bubbles and corrections. Yet overall, the system's increased agility will mitigate some inefficiencies more quickly, re-routing resources or focus as soon as signals shift. Overall, we can expect greater efficiency and rapidity, but with more frequent and faster cycles.
Fewer Gatekeepers: As friction decreases, existing structures that once decided which projects "deserved" funding or attention (e.g., large firms, centralized R&D departments) might lose power. Instead, crowds, influencers, or even AI-based recommendation systems could direct resources to new ideas. While this indeed risks ephemeral booms and busts, it can also unlock creativity from previously excluded communities. As the role of gatekeepers diminishes, we will see increasingly rapid disruption. This will primarily lead to greater diversity and efficiency, but it will also create significant economic losers and potentially result in local maxima driven by crowd-driven dynamics.
Impact on Physical Industries: Not every industry can operate at the same pace as pure software. Physical supply chains, real estate, and heavily regulated sectors operate under intrinsic real-world constraints that inherently limit their flexibility and speed. However, although "atoms aren't bits"—and physical constraints remain—much of the knowledge and coordination overhead in real-world industries can be drastically reduced by turning processes into flexible, AI-driven software. So much of what constrains atomic progress and atomic programs is actually in the world of bits. Consider the percent of people / capital at a given industrial company focused on "building the thing" versus regulators, lawyers, processes etc. Historically bits really have constrained the atoms in many cases. Malleable software will significantly streamline numerous informational and transactional elements within these industries, including inventory management, on-demand production scheduling, logistics optimization, and asset utilization. This digital orchestration will substantially reduce operational friction, enabling these sectors to progressively align more closely with the agility and responsiveness characteristic of digital markets. As such, even in non-purely software industries, from manufacturing lines to logistics, the cost of retooling or pivoting will decrease. Nevertheless, tangible physical constraints will persist, and thus transformations in these sectors won't be instantaneous or uniformly disruptive. Instead, changes will manifest incrementally but persistently, amplifying the competitive advantage of companies adept at integrating malleable software into their operational and strategic frameworks.
Rethinking Investment, Valuation, and Competition
In such a malleable environment, venture capital and other forms of funding will evolve:
Shorter Time Horizons: With product iterations measured in days or hours rather than months, investors may see returns or failures far more quickly than the classic startup arc of a multi-year runway.
Instant Market Feedback: "Product-market fit" may emerge or vanish in real-time, as user modifications or AI-driven customizations shift usage patterns. Market signals become constant and micro-targeted. Investors will need to respond to opportunities even in private markets at near real-time.
Reimagined Competition: Incumbents face threats not just from well-funded startups, but from creative individuals or microteams launching viral solutions overnight. Likewise, protective moats rely less on code complexity and more on brand, network effects, or proprietary data. Investors will need to consider new competitive dynamics. The opportunity will be greater, but the competition fiercer.
Commodification of Software: Many "software suppliers" have historically used "value based pricing" to ensure significant producer surplus and significant value capture because the thing they are producing is fundamentally difficult to produce and difficult to substitute. However, thanks to malleable software and AI, substitution will increase for a broader range of software products as it becomes less difficult to produce them. As the supply of software increases, its value should decrease. As such we will likely see a massive shift in the economy where much software suddenly does become more cost-driven from a pricing and value capture perspective as many software companies find themselves selling a less cornered and differentiated resource. The most valuable software and software companies will instead sell bundles of software and some other scarce resource, capability or input, not just code.
Atoms and Bits: As the cost of creating and adapting software approaches zero, even as the marginal cost of running software increases, the relative value of complementary physical goods and infrastructure rises significantly. This dynamic enhances the attractiveness of traditional, physically-based industries as investment opportunities, especially when these assets can be leveraged through software-driven improvements. Businesses that effectively utilize malleable software to enhance operational efficiency, responsiveness, and scalability will significantly outperform competitors who remain reliant on rigid, traditional models. This competitive disparity will drive significant market disruptions and foster investment interest in companies that demonstrate software-driven optimization. Unlike pure software businesses, however, these enterprises will maintain a more durable moat, even as the cost of software creation declines, thanks to their integration into the real world. Early signs of this shift are already apparent. For instance, startups such as Cabana, which employs AI-driven, malleable software to aggregate and efficiently manage traditionally fragmented, manual sectors—like residential pool cleaning—illustrate how digitized operational models can create scalable, venture-backed enterprises out of historically small-scale, physical-service businesses. As these trends intensify, traditional sectors integrating malleable software will become increasingly compelling investment opportunities, reshaping investor priorities and market dynamics across a broad array of industries.
Ultimately, malleable software stands to make economic activity more liquid, adaptable, and varied. We won't eliminate human error, hype, or real-world constraints, but we will see a faster-moving environment where resources—time, capital, and attention—flow more readily toward promising ideas. Over time, as friction continues to drop, industries that have historically moved slowly may find themselves operating with a speed and fluidity reminiscent of modern financial markets. And while that doesn't come without significant disruption, costs, and challenges, it does point toward a future shaped by widespread creativity, lower transaction costs, and a market-like dynamic for how work and innovation get done.
TL;DR
Malleable software, powered by AI, may be as transformative, if not more fundamentally disruptive, to the modern economy as the personal computer was in the 1980s or the smartphone in the 2010s. By giving everyday users, creative enthusiasts, and enterprises alike the ability to reshape digital experiences in real time while encoding more of the world into machine readable formats, it broadens who can build, innovate, and scale, thereby blurring the line between consumer and developer and democratizing creation both in and outside of firms.
Consumers gain the freedom to personalize and build applications at a fundamental level, remixing and extending features without specialized coding knowledge.
Enterprises become cybernetic and highly adaptive, relying less on fixed hierarchies or rigid processes. Through malleable software and AI agents, they can lower transaction costs, reduce time-to-market, and reorganize workflows at machine speed.
Investors and Markets will see capital flowing faster to teams that can execute rapidly. Valuations may shift toward brands, communities, and data networks, rather than raw technical assets. Meanwhile, market structures will increasingly reward nimble creators, microteams, and niche experts who successfully remix or augment existing software.
Developers and Platform Makers will focus on fostering ecosystems where the masses can safely experiment and remix, balancing security, IP concerns, and open collaboration.
By reducing friction and streamlining the software creation process, malleable software has the potential to reshape how we create and work. While it won't eliminate real-world constraints—especially in physical industries—it will accelerate the flow of ideas and resources, prompting a world that feels more fluid, adaptive, and market-driven. Ultimately, the ability to respond to genuine needs, harness creativity, and thrive in an environment of continuous iteration will overshadow most other skills, ushering in a new, more democratic era where taste, insight, and community rule.
Note: Thanks to Sebastian Park for helping edit and shape this essay. Thanks to Akshay Krishnaswamy for the chats that helped refine it. Thanks to Peter Wilczynski for the original car ride discussion that inspired it, the insight on program types and how to connect the various ideas together. This essay is very much a piece of malleable work, remixed from the building blocks of others including Geoffrey Litt, Benn Stancil, Bret Taylor and more, apologies if I missed anyone.
exceptional essay