The Future of AI Development

Software is collapsing into three things: a chat window, the data behind it, and the integrations that connect the two. In 2026, almost everything digital is already built with AI, and the products we grew up with — standalone apps, websites, browsers — are starting to look like transitional forms.

This is not a story about AI replacing developers. It is a story about where value moves when building software stops being the hard part. Below are the forces driving that shift, and what I expect it to produce over the next several years.

The forces at play

Five forces are pushing in the same direction at once.

AI is now the default way to build. Code, content, design and documentation are increasingly produced with AI tools. When creation is cheap, the thing being created stops being scarce — and stops being where the money is.

SaaS is dying — but only in its old shape. The classic model of one product solving one narrow problem is losing its edge. A single feature is easy to replicate in an afternoon. What remains valuable is synergy: how well a service plugs into everything else the user already relies on.

Users have learned to expect a single window. Consumers have grown used to typing a request into one chat and getting it done. Every additional app to open, account to create, or interface to learn now feels like friction.

Data is concentrated. The major AI providers hold more data than anyone else, and they gather more with every interaction. That advantage compounds.

Compute is concentrated. Whoever owns the infrastructure owns the processing power, and with it, the ability to train and serve the models everyone else depends on.

Forecast 1: The interface collapses into a chat window

Applications as we know them will disappear. In their place: a chat window with additional functions attached. The user states what they want; the AI decides which services to call.

The interfaces of major software will be radically simplified. Dense dashboards, nested menus and onboarding tours exist because humans had to drive the software themselves. When an AI drives it, most of that surface becomes unnecessary.

Websites will follow. Navigating pages, cookie banners and search results is an inconvenient experience compared with simply asking. Browsers will transform accordingly — from tools for visiting pages into AI chat windows that fetch, read and act on the user’s behalf.

Forecast 2: Software becomes data and integrations

Software will be mostly data and integrations. There will be less and less user-facing software, and more and more AI-facing interfaces whose job is to supply data to models.

Major non-infrastructure services will become context providers — sources of knowledge and capability that every other piece of software connects to. Most will be paid; some will be free, usually as a loss leader or a public good.

The economics of running software will shift with it. Instead of paying for hosting, people and companies will buy two things: context (access to the data and services their AI needs) and inference (the processing power to use it). The server you rent matters less than the context you can reach and the tokens you can afford.

Forecast 3: Data and hardware become the new moats

Starting a new software project without data will be very hard. Writing the code is no longer the barrier; having something worth serving is. New entrants will need either proprietary data or a credible way to aggregate it — and incumbents will guard both.

Hardware ownership will split the market in two. Those who own real hardware will run their own models, on their own terms, with their own data kept private. Those who don’t will use whatever the market offers, at the price and under the rules the providers set.

Trust will become infrastructure. As AI-generated material floods every channel, trusted data will be marked, signed and secured. Access to sensitive or high-value data will increasingly require full identity verification — the anonymous web will shrink to the parts nobody needs to trust.

Forecast 4: Security, code and content are redefined

Security gets rebuilt from scratch. AI can now find vulnerabilities faster than humans can patch them. Anything legacy that hasn’t been checked with AI will be considered unusable or unsafe. Old codebases will face a simple choice: audit and harden, or retire.

Open source as a movement becomes obsolete — because nearly everything becomes open. When most software is a thin gateway to context APIs, the code itself is cheap, reproducible and not worth hiding. Openness stops being an ideological stance and becomes the natural state of things. The value sits behind the API, not in the repository.

Proprietary software moves up in complexity. What stays closed will be the hard part: systems that orchestrate multiple AI models, process large volumes of data and guarantee quality across all of it. Fewer proprietary products, but each far more complex than today’s.

Content becomes AI-made by default. Around 90% of publicly available content will be created with the help of AI tools. Human-made work won’t disappear, but it will need provenance to stand out — which loops back to marked, trusted data.

Forecast 5: Hardware will become more closed

As software opens up, hardware moves the other way. When code is cheap and replicable, the physical machines that run models become the scarce asset, and scarce assets get locked down.

Supply concentrates. The capacity to produce AI accelerators sits with a handful of chip designers and fabs. Their output goes first to hyperscalers and model providers on long-term contracts. Smaller buyers get what is left, later and at higher prices.

Chips get tied to stacks. Accelerators increasingly ship bound to their vendor’s drivers, compilers and runtime. Moving a workload to different hardware means rewriting the stack beneath it, so the choice of chip becomes a long-term commitment.

Consumer devices close up. Phones, laptops and PCs will ship with on-device AI tied to the manufacturer’s own models and ecosystem. The neural processor in your machine will run what the vendor permits, not whatever you want to load onto it.

Politics adds another layer. Export controls and national-security rules already decide who can buy advanced compute. Expect more of that, not less. Access to hardware becomes partly a question of jurisdiction.

The result reinforces the split described in Forecast 3. Owning capable, unrestricted hardware becomes a privilege rather than a default. That makes it a strategic decision for any company that wants independence from the major providers. Open hardware efforts and second-hand compute markets will grow as a counterweight, but they will trail the frontier.

Forecast 6: Small businesses

Small businesses gain the most leverage and face the most dependency. AI lets one person run what used to take a team, but it also puts a new gatekeeper between them and their customers.

Being found replaces being visited. When customers ask a chat window instead of browsing, a small business’s website matters less than its data. Opening hours, prices, stock, availability and booking must be machine-readable and reachable through an API or a context provider. A business the AI can’t read effectively doesn’t exist.

A new gatekeeper takes its cut. The AI providers decide which plumber, café or accountant gets recommended. Expect “AI visibility” to replace SEO, and expect paid placement inside AI answers to follow, as it did with search ads. Small businesses will pay to be suggested, on terms they don’t control.

The back office mostly disappears. Bookkeeping, invoicing, marketing copy, customer support and basic legal paperwork will be handled by AI at a fraction of today’s cost. A two-person business will operate with the capabilities of a twenty-person one. More people will start businesses because the overhead barrier drops.

Small software companies get squeezed hardest. Niche SaaS tools that solve one narrow problem will be absorbed into the chat window or rebuilt in an afternoon. The ones that survive will own a specific dataset, a regulated niche or deep integration into an industry’s workflow.

Human and local work gains value. As digital output becomes abundant and AI-made, physical services, craft, personal relationships and verified human work become the differentiators. Trust and provenance will be things customers pay a premium for.

Costs shift from fixed to metered. Instead of paying for software licenses and staff, small businesses will pay per use for context and inference. That lowers the entry cost but creates exposure to provider price changes and lock-in.

The likely outcome is polarization. A large number of very lean micro-businesses will thrive on AI leverage, while mid-sized firms built on headcount and simple software margins come under pressure. The small businesses that win will own their customer relationships and their data, rather than renting both from a platform.

What this means for builders

The question for anyone building software is no longer “what app should we make?” It is “what context do we own, and who will connect to it?”

A second question follows close behind: how do you stay independent when a few providers control both the hardware and the window customers find you through?

• Own data, not screens. A unique, trusted dataset will outlast any interface built on top of it.
• Design for AI consumers first. Clean APIs, clear schemas and reliable context endpoints matter more than pixel-perfect UI.
• Treat integration as the product. The winners will be the services that slot into the single chat window most easily.
• Audit what you already run. Unchecked legacy code is a liability that grows every month.
• Decide where you stand on compute. Hardware is closing up, and capacity goes to the biggest buyers first. Secure the compute you need early, avoid locking into a single vendor’s stack where you can, and treat independence as a deliberate investment rather than a default.
• Plan for metered costs. Context and inference are billed per use. Build provider price changes into your margins and keep a fallback provider ready.
• Make your business readable by AI. Publish prices, availability and booking as structured data. AI visibility is the new SEO.
• Own the customer relationship. Don’t let the chat platform become the only path to your customers. Keep direct contacts and first-party data.
• Sell what AI can’t fake. As digital output becomes abundant, human, local and verified work commands a premium.

The era of software as a destination is ending. The next one belongs to whoever supplies the context the AI needs.