AI for software developers

There are many tools for coding with AI, but only some have a very low entry barrier and can be tried without too much friction.

This article focuses on solutions we can use right now, at low cost, and without getting into custom deployments or self-hosted infrastructure.

What AI brings to development

Used well, AI can speed up tasks like:

• Autocomplete and refactoring
• Onboarding processes and repository navigation
• Test generation/modification, migrations
• Initial pull/merge request review

It does not replace humans or their judgment/technical knowledge, but it does reduce a lot of repetitive work.

Quick recommendations

Quickly, these are my recommendations:

Core concepts

There are two key pieces in using AI for coding: inference providers and interfaces.

Inference providers

These are services that provide the compute needed to run AI models and usually also the models themselves.

For example, OpenAI is an inference provider because it lets you use its models and infrastructure, either through API or its own products.

A rented GPU on Vast.aiis another option, but here we’re talking about infrastructure where you’d have to deploy your own inference service. It’s more complex and out of scope for this article.

Interfaces

These are the applications we use to interact with inference providers.

They can be:

• Traditional IDEs with added AI capabilities, such as VS Code with extensions
• Agent-oriented CLIs, such as OpenCode or Claude Code
• Traditional UI applications like Codex
• Integrated review systems in GitHub or GitLab

ChatGPT vs. these coding tools

The key difference between using a web chatbot (like ChatGPT) and using one of these specialized tools is context.

It is not practical to paste a whole large repository into a chat: there are context limits, cost, latency, and too much noise for the model. In contrast, specialized tools use system capabilities, such as file reading, search, terminal, or Git, to query only what they need at each moment.

That improves two things:

• Response quality
• Ability to make changes in the real project code

IDEs and development environments

There are many IDEs with generative AI capabilities. Many of the most popular options are based on VS Code or its ecosystem, but those are not the only ones.

Some alternatives:

• Zed
• GitHub Copilot for IntelliJ IDEA
• PyCharm
• Microsoft Visual Studio

VS Code

In VS Code, AI is integrated mainly through extensions, both official and community-made.

GitHub Copilot

What it is:

• The most widespread option in enterprise environments
• Combines chat, autocomplete, and assisted editing inside the editor

Strengths:

• Easy adoption if you already work with GitHub and Microsoft
• Strong autocomplete experience
• Good fit as an entry point for teams

Limitations:

• The model catalog and provider flexibility are more limited than in more open tools
• By default it is a more closed service than alternatives like OpenCode or Cline

When I would choose it:

• If a company wants to adopt a solution like this and do it relatively fast

Reference: GitHub Copilot

Kilo Code / Cline

What they are:

• Open source projects with an IDE extension and also a CLI
• They separate the interface from the inference provider

Strengths:

• They allow using many different providers and models
• They provide more control over cost, quality, and configuration
• They are good options for BYOK

Limitations:

• They require more configuration
• The experience usually depends more on how provider, model, and permissions are tuned

When I would choose them:

• When you want flexibility, to test different models, or for personal projects without leaving VS Code

References: Kilo Codeand Cline

Cursor

What it is:

• A VS Code fork focused on AI from day one

Strengths:

• Very polished experience
• Autocomplete and assisted editing
• Agent-oriented workflow

Limitations:

• It is a more closed product than VS Code, some extensions do not work in Cursor
• Some advanced features depend on your paid plan

When I would choose it:

• When more serious, agent-oriented work is needed while keeping the strengths of VS Code

Besides the IDE, Cursor offers complementary products such as automated change review and cloud agents. It is worth trying because it sets a pretty good benchmark for how far user experience can go.

Reference: Cursor

AI-native environments and agent-oriented CLIs

This section includes tools designed around agent workflows, not just as an added editor extension.

OpenAI Codex

What it is:

• An agent-centered development suite with its own interface

Strengths:

• Very guided experience
• Convenient if you already work in the OpenAI ecosystem
• Good handling of parallel tasks and per-project context

Limitations:

• It is centered on OpenAI models
• It is less flexible if you want to mix providers or external models

When I would choose it:

• If you already pay for ChatGPT and want an integrated and simple experience

Reference: OpenAI Codex

Claude Code

What it is:

• An agent-oriented development CLI, very focused on terminal-first work

Strengths:

• Intuitive for people very comfortable with the terminal
• Can also be used from the Claude desktop app

Limitations:

• Non-Enterprise users can only use models provided by Anthropic

When I would choose it:

• If you prefer terminal workflows, value simplicity, and want to work with Anthropic models

Note: today I would not choose this option with Cursor and OpenCode available.

Reference: Claude Code

OpenCode

What it is:

• An open source project that started as a CLI, though now it also has a desktop app and IDE extension

Strengths:

• Very flexible across providers and models
• It often provides models for free
• Good option to compare models without changing tools

Limitations:

• It requires a bit more judgment to choose provider and model
• The experience can vary quite a lot depending on configuration

When I would choose it:

• When you want maximum flexibility and to optimize cost without being tied to a single ecosystem

Interesting options inside OpenCode:

• OpenCode Go: low-cost monthly plan with access to several selected models
• BYOK (Bring Your Own Key) with OpenRouter: useful if you want access to many models from a single integration point
• Direct providers: good option if you already have accounts or contracts with specific providers (OpenAI, Alibaba, Moonshot, etc.)

Reference: OpenCode

Comparison table