July 2025
From: Brian, Tobias, and Gaby
Subject: The start of the developer AI renaissance
Welcome back to the BT&G newsletter. We hope your summer is going swimmingly.
While VCs try to catch a summer breather, the developer AI gold rush is on. Writing code is AI’s most powerful use case – second only to chat – and we’re seeing a mad dash for mindshare among incumbents and challengers alike. The shift is seismic. If software is eating the world, AI is eating the software engineer.
Copilot quietly launched on JetBrains in 2021, but ChatGPT made it clear: code automation is here. Cursor, founded by MIT students in 2023, is now valued at $9B. Cognition’s Devin debuted in March 2024 and is reportedly raising at a $10B valuation – hot on the heels of its buzzy Windsurf acquisition, which came days after Google’s acqui-hire of the exec team.
Meanwhile, pre-ChatGPT players like Warp, Graphite, and Tabnine are retooling and gaining steam. Some had strong distribution, but the tech was leapfrogged overnight. Founders are rising to the moment. As Warp CEO Zach Lloyd puts it: “It’s still early innings in codegen. The first wave focused on autocomplete. Now, it’s all about agentic development – and the right tool won’t look like a terminal or IDE, but something purpose-built for telling your computer what to do.”
We firmly believe this type of wholesale reimagining of software development is coming with the AI revolution.
The core driver here is model performance. Engineers tell us the tipping point for agentic capabilities came with Claude Sonnet 3.7 in February. Anthropic continues to top leaderboards, especially on SWE-Bench, where Claude Sonnet 4 leads. A hybrid reasoning model, Sonnet 4 delivers low latency and deep reasoning across massive codebases, thanks to its massive (200k tokens) context window. The next generation of dev tools will be defined by how well they harness these breakthroughs.
In this newsletter, we explore how AI is blowing open the developer tools stack and changing not just how software is built, but who gets to build it. AI-native tools are redefining the role of the engineer, rewriting business models, and rebuilding the software stack from the ground up.
Let’s dive in.
The business of dev tools in an AI-native world
AI-native dev tools have a conundrum. They face fierce competition and structural challenges that compound one another. On one hand, code generation requires massive compute and is expensive; the more you grow, the more you burn. On the other hand, lock-in is hard. Startups are paying for growth, burning cash, only to find their users churn. Switching costs are low and the developer appetite to experiment is high.
The companies are attacking these challenges head-on by leaning into workflows that generate proprietary data, building tight integrations across the stack, and embedding themselves into daily developer habits. Some are wrapping foundational models with custom orchestration layers to manage cost and latency more effectively, or building their own models to avoid the API tax. Others are focusing on niche use cases where performance trumps switching friction. Top performers are treating infrastructure not just as a cost center, but as a strategic advantage, and they’re marketing like consumer businesses.
The reinvention into the AI-native era
For dev tools companies that launched before October 2022, the arrival of generative AI made one thing clear: adapt or be left behind. It became abundantly clear that a rethinking was required to meet the moment. The transformation to become AI native has been successful for a select number of dev tools that existed before ChatGPT came on the scene:
The most successful players here won’t just layer AI onto legacy tools, they’ll rethink their products from the ground up. We’ve heard of customer backlash and team resistance, but we’ve also seen momentum build around the company once users and employees see their first “AI aha” moment with the new product.
Software engineering for all builders
Software engineers have a long history of being revered in startup land. We don’t think that will change, but who can be an engineer will change. The junior software engineering market has contracted sharply, replaced in many cases by AI agents that can perform their work faster and at scale. But more companies are seeing the need for AI-builders who can vibe-code but may lack formal training or deep experience. Startups are hiring for engineers not with hand written algorithm tests, but with questions around how well you use AI tools. In major businesses like Microsoft and Google, AI is now writing 30%+ of all code.
This isn’t just a technical shift – it’s cultural. For decades, software engineers were the most valuable people at startups – highly skilled, irreplaceable, and richly compensated. Engineering was both an art and a science, reserved for trained experts. However, AI is making the skills commoditized and consumerized, which is a real shake up to the traditional ethos for startups, and the startup economy more broadly. As the Lovable CEO said, “99% of the world's best ideas are trapped in the heads of people who can't code”. It might even be hard to call the category “dev tools” in the future, when most of the users aren’t traditional developers.
Those who build software in the future will fall into three camps, starting with the most abstract:
If this is true, then we’re just at the beginning of what developer tools will really look like in the age of AI. The underlying infrastructure is shifting rapidly, redefining roles within the organization and how software is developed entirely. This requires us to rethink collaboration – as Mehdi Jamei, CEO of Veris AI told us, “if you build a layer that can enable PM-Dev collaboration on agentic workflows and build prod-ready systems, that'd be a huge unlock.” The bridge between the researchers, LLM wranglers, and systems architects will look differently than what exists today.
On top of this, the reality is that the software stack of the future likely won’t be built for humans. It will be optimized for agents to read/write/modify codebases and surrounding infrastructure. As the tools become more comprehensive in what they can do, the market of end users will also expand.
Where we’re going and what we’re looking for
Just as accounting and mathematics changed entirely when the calculator, then Excel, came out, the tools that are built today will require a new set of skills for their users. Cris Dobbins, CPXO and advisor to many AI dev tools startups, said it best – we need to be “rethinking software as a shared space where everyone contributes meaningfully. The future of development won’t be modality-specific. It’ll be collaboration by default, with AI assisting contextually across every layer, from mapping user flows to editing components to running agentic A/B tests. We’ll build better, together, alongside AI.” We believe this transition will create not just better dev tools (which will be used by far more people than just devs), but entirely new job roles, new business models, and a new definition of what it means to develop software.
And there is a Figma-scale business being built right now. Some ideas we’re looking for:
As always, we’re eager to meet folks building in these spaces and hear about how you are thinking about the future of developer AI. And many thanks to the builders who offered their thoughts, including Ivan Burazin, Mehdi Jamei, Cris Dobbins, Marcus Moretti, and Zach Lloyd.
Best,
BT&G