Predictions for 2026: 5 Topics

We’re in the midst of the largest shift in how the web is built, who is building it and what they are building since the creation of the web. Here are my predictions for the year that just started.

True Computer Literacy

I was born in 1976, at the birth of personal computing. I witnessed the sweeping change of generations that had to grow into computers and generations growing up with computers, and saw “computer literacy” go from a job search advantage to a requirement in just about any profession.

Literacy is “the ability to read and write”. In contrast, computer literacy only covers the ability to use computers and software – not the ability to write your own software and build your own programs.

AI coding agents will drive a similar wave of generations having to grow into True Computer Literacy (meaning not just consuming software, but writing it as well) and generations growing up with this.

Claude Code with Opus 4.5 has reached a point where lots of skilled programmers realize that they are suddenly doing the majority of their work without an IDE or code editor open. From Gergely Orosz, over DHH to Jaana Dogan, we’ve seen similar realizations that something fundamental has changed now and that software development will never be the same.

2025 was the year where tools like Bolt.new, Lovable, and Repl.it surged in popularity and revenue as they opened the doors to building software for millions and millions of new users. But still with a sense that there was a fairly low ceiling for what a non-coder could build before needing something like vibecodefixers.com to bring on board professional programmers to help them out.

2026 will be the year where that ceiling will be broken. The skills involved in building real software for business purposes through prompting will be the True Computer Literacy. It will be a differentiated advantage in lots of jobs. And once you start building real-world value-creating software at work – with AI Agents writing the code – You are now a developer.

Claude Agent SDK

Claude Code launched into GA in May 2025, after having started as an internal engineering experiment at Anthropic by Boris Cherny.

Today, Claude Code is a $1B+ business line for Anthropic, and as I’ve argued before this type of CLI-based Agent turned out to be the humanoid robot form factor of cyberspace. Claude Code was quickly followed by Gemini CLI and Codex CLI from Google and Open AI.

Claude Code is one of the first true AI Agents to be fully adopted across industry for production purposes. It’s not a “copilot” that works alongside a programmer inside an IDE or a “workflow” with defined steps. It’s a true agent that has full access to the computer it’s running on and a fluid set of tools and skills that it can use to do autonomous work towards a goal given in a prompt.

Anthropic’s Agent SDK is essentially the Claude Code agent as a library for Typescript or Python. It makes it trivial to build agents for other purposes than coding using the same agent harness. You define the tools and skills and the right sandboxed environment and you can now prompt an agent to work towards a goal.

Today, most enterprise “agents” are typically really just durable workflows with LLM calls at specific steps. 2026 will be the year where we start seeing broad application of true autonomous agents based on the patterns embodied by Anthropic’s Agent SDK. We’ll see much longer truly autonomous agent runs without a human in the loop, and they’ll navigate more complex problems through skills and tool use.

For these kinds of agents, a large ecosystem will grow up around Agent Skills. They will be more relevant than MCP servers since they fit better with the UNIX model of simple tools that the CLI-based agents can compose together as they see fit.

AX, Agent Experience

I first wrote about Agent Experience about a year ago at the start of 2025 as Netlify was fully changing our product direction to be the best web platform for agents to work with.

The term quickly resonated amongst dev tools builders and founders, since 2025 was also the year of adoption for autonomous coding agents. The term quickly gained prominence in the tooling and infra space ranging from web platforms like ours, over open source frameworks like Tanstack or Vite, data platforms like Supabase, Neon or MongoDB, auth providers like Clerk, Styck, WorkOS, or Auth0, to emerging sandboxing providers like Daytona.

With the fast rise of coding agents, it was evident for any forward-looking dev tool founder that tools that failed to work well in this new paradigm of building software would simply become irrelevant. And also evident that tools like Netlify who fit well into this agent-driven development workflow could massively expand their total addressable market as the web developer segment expands from about 17 million professional JS developers to 3 Billion Developers.

In 2026, as Anthropic’s Agent SDK and similar offerings from foundational model providers allow companies to apply truly autonomous agents to larger and larger parts of the work they are doing in the digital world, the more we’ll see AX start to become a key concern not just for dev tools, but for any kind of digital experience. Similar to how UX is an indispensable discipline today.

Pipelines and Sandboxes

Writing and working with code is now an order of magnitude faster than it’s ever been before. The new primary constraint on building software will become the pipelines driving a new piece of code through verification and testing to production. The workflows allowing rapid iteration and frictionless operation of production software.

Speeding up this part of the process of building for the web has always been Netlify’s core value proposition: hooking into the git workflow, creating deploy previews with their own environments for each pull request or branch, offering instant rollbacks when something goes wrong, etc.

While the level of throughput this kind of workflow offers has been important for fast-moving teams for a long time, coding agents are making it absolutely essential. Without fast pipelines and isolated environments, the acceleration at the coding step just starts creating congestion and chaos throughout the rest of the system.

Code reviews are another bottleneck that’s quickly emerging based on the current standard workflows around Pull Requests and reviews. Requiring human developers to spend most of their time doing manual code reviews of massive amounts of machine-generated code will end up being backwards. But without thorough code reviews, we’ll need the ability to do evals and in-depth testing of all functionality of our applications at a much faster pace than we’ve been doing them so far.

This will require sandboxes for each unit of change we want to bring to production. Isolated environments that we can run automated test suites against. Controlled rollouts into the real world with built-in observability so agents can monitor and catch any regressions or anomalies that reach production.

2026 will be a year where the most efficient pipelines from code to production will determine how fast you can build.

The beginning of the end for syntax and structure

Computers and software are fundamentally centered around data structures and algorithms that manipulate them. This will never change at the lower levels of computing, but it’s been a leaky abstraction that’s all around us today.

The way we interact with applications have generally been through “forms” with structured fields with clear data types and validations. We have content management systems to handle the input and operations of structured data that websites, e-commerce platforms, or applications are built around.

We write software through programs with strict syntax where one wrongly placed comma will completely stop the software from compiling or working, and write tests or verifications in the same way.

LLMs are radically different because they really don’t care that much about formal structured data. They work fine with the same loose structure that humans work with and can even process images and non-textual data.

Starting to build our platforms around this will fundamentally change how we work with software and computers. Managing structured content mostly becomes moot, where searching and querying loosely structured data becomes much more important.

Interchange data formats, Content Management, programming languages, will all start decreasing in relevance.

We’ve already seen examples like Lee Rob removing the CMS from cursor.com to make it easier for code agents to work with or Wayne Sutton’s building a simple markdown CMS for the same reason, but 2026 will be the year that starts to fundamentally alter our apps, the web, and the tools and platforms we build with.