The battle for AI supremacy is intensifying.

This is The Stepback, a weekly newsletter diving into a key story from the tech landscape. For insights on the AI coding and vibe-coding trends, keep an eye on developments in the field. The Stepback lands in subscribers’ inboxes at 8 AM ET.

How it started
The journey of AI in coding began long before it became a buzzword. Back in spring 2021, Microsoft launched GitHub Copilot, a tool developed in collaboration with a nonprofit, which assisted developers by suggesting code snippets as they typed. Although it was just a “restricted technical preview” and not particularly effective, over a million developers were eager to give it a shot.

Large language models (LLMs) seemed destined to streamline software development. Given that most code is structured and coding languages are well-documented, there’s a wealth of online code available for training models. Unlike other types of information from LLMs, the quality of code can be verified by running it. Initially, companies thought LLMs could enhance coding speed by predicting the next word, similar to Google’s autocomplete. However, the ultimate goal was to have these models handle coding tasks autonomously.

For years, the tech industry has explored “low code” and “no code” solutions, aiming to empower users to create software without getting lost in complex settings and menus. Early attempts were often clunky, with tools like Zapier and Apple Shortcuts serving as intricate workflow builders, while platforms like Notion and Airtable offered flexibility at the cost of usability.

Even back then, the potential for AI coding tools as a lucrative business was clear. Developers are costly, and product development is time-consuming. Any tool that could reduce the need for developers or boost their productivity would be an easy sell to software companies. If the technology worked, the products would practically market themselves. Startups like Cursor and Windsurf attracted significant funding to develop AI coding solutions, while major players began crafting new tools for developers.

Initially, AI coding tools were unreliable, often requiring thorough checks even for simple code completions. In late 2023, programmer Simon Willison referred to LLMs as “weird coding interns,” pondering whether they would enhance coder capabilities or eventually replace them. By early 2025, Anthropic introduced Claude Code, raising the stakes for many in the industry.

How it’s going
In late 2025, Anthropic launched Opus 4.5, the latest iteration of its Claude LLM. While it was touted as the best version yet, it didn’t signify a groundbreaking leap in AI tech. However, as developers began experimenting with Claude Code during the holiday season, a consensus emerged: it worked remarkably well. The tool that once required careful prompting and review could now transform a few sentences into a functional prototype. Boris Cherny, the creator of Claude Code, shared that he had AI generate 100 percent of his code, expressing surprise at the tool’s capabilities. Claude Code quickly gained traction in the software community.

While Claude Code captured attention, competitors were not far behind. OpenAI’s Codex, released shortly after Claude Code, received updates and became a formidable tool for coding. Google also introduced a command line interface for its Gemini model, enhancing its AI Studio app with coding features.

AI coding is rapidly becoming a mainstream application of AI, and it’s shaping up to be a significant business opportunity. The success of Claude Code coincided with a surge in revenue for Anthropic, prompting OpenAI executives to refocus their efforts on competing with it. Both companies are reportedly eyeing public offerings this year, needing to demonstrate value for the substantial investments they’ve received. It seems that writing code is becoming everyone’s top priority.

This trend appears to be a reasonable bet. Companies in Silicon Valley are witnessing employees vying to utilize the most tokens, leveraging GPU access as a recruitment tool, and openly discussing their AI expenses. Nvidia’s CEO recently expressed concern for any highly paid engineer not spending $250,000 annually on AI tokens. Despite fears that AI coding tools could threaten developers’ careers, the rush to adopt these technologies is palpable. A 2025 study revealed that 98 percent of respondents used AI coding tools several times a week.

It’s not just developers getting in on the action. In February 2025, AI expert Andrej Karpathy introduced the term “vibe coding.” He described a process where he was building a project or web app without traditional coding—just seeing, saying, running, and copying/pasting, and it mostly worked.

It’s unfortunate that Karpathy didn’t coin a catchier term, as “vibe coding” has taken off. This phenomenon allows individuals who may not know how to code to create functional software through prompts. For many, who might have otherwise created slide decks or mockups, a basic prototype suffices, and these tools have proven capable of delivering just that. However, vibe coding carries risks, including potential issues from faulty code and privacy concerns when granting these tools access to personal data. Trusting a system is one thing when you can verify its output; it’s another when you can’t understand its workings.

What happens next
The software developer landscape is just beginning to shift. Companies in Silicon Valley are laying off thousands, often citing AI as a factor. Block CEO Jack Dorsey noted that a smaller team using their tools could achieve more, emphasizing the rapid advancement of intelligence tool capabilities. While AI may be a convenient excuse for pandemic-era overhiring, the tech industry is clearly leaning into AI to boost productivity and reduce workforce size.

As AI coding tools evolve, they could reshape the software industry. Why pay for someone else’s software when Claude Code can create it to your specifications? Some are dubbing this the “SaaSpocalypse,” predicting a fundamental reevaluation of software value. Others foresee a new wave of startups offering AI-driven solutions. Still, some believe the concerns are exaggerated and that established companies will remain stable. Regardless, the software industry, which has reached unprecedented heights, now feels precarious to many.

On the other end of the spectrum are vibe coders. For most, even the simplest AI coding tools can be overwhelming. They often require reading code, Terminal access, and answering complex questions that many users may not know how to tackle. AI coding still faces numerous bugs, privacy issues, and vulnerabilities that bad actors could exploit.

With initiatives like Claude Cowork, Anthropic is exploring ways to make Claude Code’s technology more user-friendly. Users can grant it access to files on their computers and let it handle tasks. Other products, like Perplexity Computer, are investigating whether users would allow LLMs access to their devices for organizing files, responding to messages, or even making purchases. While the underlying technology is progressing, it remains unclear how users will engage with it and whether they will embrace it.

By the way
* If you haven’t tried vibe coding yet, it’s worth exploring—seeing how these tools function is quite fascinating. Start with Claude Cowork, give it access to your Downloads folder, and watch it organize everything for you.
* Currently, most users are paying either $20 or $200 monthly for AI coding tools. OpenAI has introduced a $100 tier aimed at heavy Codex users. As these companies seek profitability, the $20 plan may not offer much value moving forward.
* OpenAI, Anthropic, and Google are all eager to get users engaged with their apps, striving to create AI super apps with coding as a core feature. Anthropic has recently attempted to restrict OpenClaw, indicating that these companies will likely continue to tighten their ecosystems to keep users within their platforms.

Read this
* We discussed the Claude Code Moment in early 2026, and the insights still resonate.
* Paul Ford wrote an insightful piece on how coders perceive AI coding, which he later discussed on a podcast. Clive Thompson also contributed an excellent article on the topic.
* The vibe-coding subreddit on Reddit offers a captivating glimpse into what people are creating and how. The running joke is that everyone seems to be building a habit tracker… because, well, everyone is building a habit tracker.