Автор Kotlin про будущее языков программирования. Андрей Бреслав, Global Talent Club

Summary

CodeSpeak and the Future of Coding

Kotlin creator Andrey Breslav describes his new startup, CodeSpeak, as an attempt to move beyond chat-based “vibe coding” toward a workflow suited to larger, industrial software projects. The core idea is that people collaborate through human-readable specifications, while AI generates and updates the conventional code behind the scenes.

How CodeSpeak Is Intended to Work

Instead of treating generated code as the main artifact, teams would work on a structured description of what the software should do. For larger systems, that description would need modules, abstraction levels, and ways to break big tasks into smaller ones. Breslav argues that expressing ideas clearly can also help people uncover gaps and contradictions in their thinking.

CodeSpeak is at an early stage: its current form is a command-line compiler that can generate an application from a Markdown document. One proposed workflow is to change the specification and have the corresponding code updated.

A Higher Level of Abstraction

Breslav places CodeSpeak in a long progression from machine code and assembly toward increasingly expressive languages. He sees it as a possible next step beyond languages such as Kotlin: machines could handle more boilerplate, while people focus on intent, system structure, and decisions.

He expects engineering skills—especially decomposing complex problems and organizing their parts—to remain important, even if reading today’s programming languages becomes unnecessary for many developers.

Current AI Coding Tools

Breslav distinguishes between tools aimed mainly at quickly creating prototypes or interfaces and tools such as Cursor, which combine an agent with code editing and can make useful changes to existing code. He considers Cursor a transitional tool: it retains the traditional-code workflow while adding AI assistance.

He expects current agents and other technologies to carry over into future systems, but is skeptical that chat-based coding alone will scale to large, complex projects.

AI’s Effect on Software Jobs

Breslav does not expect mass replacement of engineers in the near future. Instead, he predicts that the work will change and be redistributed. Lower software-development costs could make it practical to automate processes that are currently too expensive to customize, creating more demand for software.

Existing codebases will also require ongoing human support, and adoption is likely to be gradual rather than immediate.

Technical and Product Challenges

The team needs to determine how to make ambiguous natural language produce predictable software, design and test the specification language and its modules, and build suitable IDE and runtime features. Slow model responses can interrupt developers’ concentration.

Breslav also argues that AI systems often send overly simple questions to unnecessarily large models. Routing tasks to smaller models where appropriate could improve efficiency. At the time described, the team had three engineers and was hiring.

Funding and Early Users

Breslav says CodeSpeak had raised a round led by Accel, with other smaller investors involved or still finalizing. He had no prototype or pitch deck when the process began and describes the investment as largely based on his track record and the timing of strong investor interest in AI. He cautions that this is not a repeatable fundraising formula.

For adoption, he favors starting with individual developers and small, dynamic teams, then moving toward larger organizations. He wants to open-source the project’s core, but says the team had not yet decided what to release without harming the business’s commercial prospects.

Deployment and Visual Tools

Breslav says generated software needs to be runnable, so CodeSpeak will need to address deployment and related infrastructure. However, the range of DevOps and CI/CD environments is too varied to solve all at once.

He sees visual automation tools such as Integromat and Zapier as useful for accessible, limited-scale automation, but notes that diagrams become difficult to understand as systems grow. He cites UML as a cautionary example and sees visual interfaces as useful at certain scales rather than as a universal replacement for text.

Unlike predefined visual blocks, AI systems can infer a structure from a natural-language request—an advantage that also creates challenges in controlling the result.

Reviews, Guides, or Tutorials

The video is an interview and discussion, not a product review or step-by-step tutorial.

Main Speakers and Sources

• Andrey Breslav: Kotlin creator and CodeSpeak founder
• Andrey Shalamentsev: Global Talent Club founder and interviewer