When AI Can Rebuild the Software, What Are We Actually Paying For?
When AI Can Rebuild the Software, What Are We Actually Paying For?
AI-assisted development, open-source alternatives and the future of subscription software.
When AI makes it substantially cheaper to reproduce software, what are we actually paying for in a subscription?
We are beginning to see AI-assisted, open-source projects attempt to recreate familiar experiences from Microsoft Office and Adobe's creative applications.
But reproducing a feature list is not the same as reproducing enterprise trust. And that distinction may determine where the value of professional software resides in the years ahead.
If the cost of building software is falling, what prevents the value of software from falling with it?
One Development Effort, Seven Familiar Categories
The Craft projects from the Storytold GitHub organization are an interesting example of what is changing. The initiative spans seven applications — each inspired by an established commercial product — developed with AI assistance and released as open source.
Open-source alternatives are nothing new. LibreOffice, GIMP and Inkscape have challenged established products for years. What makes the Craft initiative different is the breadth of its ambition: instead of challenging one commercial application, it attempts to address several categories of software traditionally served by large, specialized vendors.
Each application is a clean-room reimplementation written entirely in Rust, dual-licensed MIT/Apache-2.0, and targeting macOS, Windows, Linux and the browser via WebAssembly. This is a useful signal of what AI-assisted development may enable: smaller efforts attempting to cover more ground than would previously have been practical. It does not establish that the applications are already equivalent to their commercial counterparts — they are evolving projects, and professional readiness must be assessed individually.
But the direction raises a question worth examining. If the cost of building software is falling, what prevents the value of software from falling with it?
The Cost of Creation Is Not the Cost of Trust
A spreadsheet that looks like Excel is not necessarily a reliable replacement for Excel. Professional users depend on formula correctness, complex workbook compatibility, charts, pivot tables, external data connections, performance and predictable behaviour when files move between applications.
A word processor must preserve formatting, pagination, references, tracked changes and document structure. Creative software must deliver precise controls, reliable rendering, colour management and predictable results. Beyond features, professional software requires accessibility, security, maintenance, recovery, documentation and support.
These capabilities are often invisible until they fail.
The cost of generating software is not the same as the cost of making software dependable.
AI can accelerate implementation, but an application still needs testing, validation, maintenance and a credible way to handle defects. For an enterprise, the relevant question is not whether a new application can demonstrate the desired features. It is whether the application can support the business process reliably, under real operating conditions, for as long as the organization needs it.
This distinction separates an interesting prototype from a viable enterprise capability — and explains why the emergence of AI-assisted alternatives should not be interpreted as the imminent end of commercial software subscriptions.
The Second Disruption: Users May Need Less Software
I use ONLYOFFICE for my office suite. I previously used Affinity Designer, but I now use AI tools to generate images. These are personal choices, not evidence that a product category has become obsolete. They illustrate a change in how I approach a task.
For documents, I need reliable editing and compatibility. For visuals accompanying my writing, AI image generation can often deliver the result I need without requiring me to perform every step of a traditional illustration workflow.
The task remains. The method changes.
AI introduces two distinct pressures on software economics. First, AI-assisted development can lower the effort required to build competing applications — challenging the cost of supplying software. Second, AI can perform tasks that previously required users to operate specialized software themselves — challenging how much of the traditional software experience users need.
A user who once needed a specialized application for a particular task may discover that an AI-assisted workflow, a simpler application or a combination of tools is sufficient. Complex work still requires precision, control and verification. But the value of a large feature set may decline for users who only need a small subset of its capabilities.
AI is not just making software easier to build. It is changing what users need software to do, what they expect to pay for, and where they expect control to reside.
Open Source Changes the Balance of Control
Subject to applicable licences, open-source software allows users to inspect, modify and redistribute source code. That creates options beyond accepting a vendor's pricing and product direction. An organization might adapt the software, contribute improvements, maintain a fork or engage a service provider to support it.
But source availability is not the same as independence. A project may depend on a small number of maintainers, external libraries or particular services the organization does not control. A permissive licence does not guarantee continued maintenance, compatibility with future document formats or timely security fixes. Open source can reduce some forms of dependency while creating others.
Which dependencies are we willing to accept, and which capabilities do we need to be able to control ourselves?
That is an Enterprise Architecture question — not simply a software procurement question. Its strategic value lies partly in the options it preserves.
Software Built for Agents
All seven Craft applications are written entirely in Rust — dual-licensed MIT/Apache-2.0, running natively on macOS, Windows, Linux and in the browser via WebAssembly. That choice is not incidental: Rust's performance and memory-safety characteristics matter for production-grade document and image processing.
But the more architecturally significant detail is what the READMEs say about agent access. DeckCraft states that "every button is also a command that the CLI and AI agents can invoke." LightCraft describes itself as "drivable end to end by AI agents over MCP." Some of the Craft applications expose command-line interfaces and Model Context Protocol (MCP) interfaces — allowing compatible tools and AI agents to invoke application functions through defined interfaces rather than relying exclusively on a person clicking through a graphical interface.
This points toward a different way of thinking about software architecture. Consider the two interaction models:
These are simplified representations, not mutually exclusive architectures. An application can support both approaches, and agent access does not automatically make it reliable or autonomous.
But the distinction matters. Imagine an AI agent gathering information, analysing a spreadsheet, generating charts, inserting them into a report and preparing a presentation for human review. The applications become components of a coordinated process rather than isolated destinations for manual work.
Software may increasingly need to serve two kinds of users: people working through interfaces and agents invoking capabilities. That creates new requirements for permissions, input validation, predictable behaviour, error handling, auditability and verification. An interface that makes a capability accessible to an agent must not automatically grant that agent unrestricted authority to use it.
The competitive unit of software may gradually shift from the standalone application to the dependable capability that can be composed into other workflows.
That shift could be more disruptive to software economics than the emergence of free alternatives alone.
What This Means for Enterprise Architecture
For Enterprise Architecture, the question is not simply whether an organization should replace Microsoft 365 or Adobe with open-source alternatives. The real question is how changing software economics affects business capabilities, application portfolios, technology dependencies, operating costs and strategic choice.
Consider an enterprise evaluating a document-processing capability. Its options might include a commercial suite, an established open-source application, a new AI-assisted alternative or a workflow in which AI performs part of the work directly. Each option has implications across seven architectural dimensions:
There is no universal winner. The architectural responsibility is to evaluate the complete operating model rather than the price of the licence in isolation.
For organizations operating in Saudi Arabia and the wider GCC, control over information, processing locations, external dependencies and regulatory obligations carries particular weight.
Applicable requirements — including relevant NCA cybersecurity controls and NDMO data-management and governance requirements — must be assessed against the organization's sector, data classification and specific regulatory obligations.
The fact that an application is open source does not automatically make it compliant. A commercial subscription does not automatically make it non-compliant. Both require an assessment of where data is processed, what telemetry or external services are involved, how access is controlled and how records are retained.
Sovereignty is not simply about owning the source code or hosting an application locally. It is about retaining meaningful control over the information, capabilities and dependencies on which the organization relies.
Three Possible Futures for Software
These scenarios are not mutually exclusive predictions. Different categories of software and different organizations may experience all three simultaneously.
Subscriptions survive, but customers increasingly expect measurable value beyond access to familiar applications.
Organizations select different tools for different needs. The result is greater choice — but also a greater need for architecture, integration and portfolio governance.
The organization may care less about which application owns each feature and more about whether the overall workflow is reliable, secure, economical and governable. Software vendors may compete not only on features but also on the quality of their APIs, agent interfaces, permission models and verification mechanisms.
The Risk of Confusing Possibility With Readiness
The excitement around AI-assisted software development can make implementation speed appear more important than it is. A demonstration may be impressive while the application remains unsuitable for critical work.
// Enterprise readiness requires more than a feature demonstration
The same discipline applies to open-source projects as to commercial products. Building a credible alternative is one achievement. Sustaining it across operating systems, hardware, file formats, changing user expectations and security requirements is another.
AI may lower the cost of reaching an initial milestone. It does not remove the need to fund and organize the work that follows.
So, What Are We Actually Paying For?
The Craft projects do not prove that subscription software is doomed. They demonstrate why software vendors should not assume that the cost of reproducing features will remain a durable barrier to competition.
When implementation becomes more accessible, the sources of value may shift toward reliability, integration, workflow quality, support, security, interoperability and the ability to evolve without losing control.
For users, that creates more choice. For vendors, it creates pressure to justify recurring payments through continuing value rather than habit or switching costs alone. For enterprises, it creates an architectural responsibility: decide which capabilities to buy, which to adopt from open-source ecosystems, which to build or adapt, and which workflows AI may transform altogether.
I will continue using tools such as ONLYOFFICE where they meet my needs, and AI image generation where it is a more effective way to accomplish the task. I do not need to replace every application to benefit from the change. That may be the most important lesson here.
AI is not just making software easier to build. It is changing what users need software to do, what they expect to pay for, and where they expect control to reside.
Comments
Post a Comment