How to choose your coding assistants
Source: Dev.to
Why It’s Harder for a Professional Developer to Use a Tool Despite the Wide Variety of Choices
Coding assistants such as Cursor, Windsurf, Claude Code, Gemini CLI, Codex, Aider, OpenCode, JetBrains AI, etc., have been making the news for the last few months. Yet, choosing the right tool can feel much harder—and more limited—for many of us than it appears.
TL;DR – OpenCode > Claude Code > Aider > Copilot > …
Not all tools are created equal. They evolve rapidly, so the examples below may become outdated soon.
1. How to Visualise the Landscape
You can plot coding assistants on a graph that shows human‑involvement vs. automation (less involvement = more automation).
| Category | Description |
|---|---|
| GitHub Copilot (early releases) | Tab‑completion, single‑line or block completions. Intent expressed via well‑named functions or comments; later added inline prompting and chat sessions. |
| Coding agents | The current state‑of‑the‑art for most developers day‑to‑day. Treat them as teammates (albeit with anterograde amnesia). |
| Parallelised / background agents | e.g., Claude Code’s sub‑agents for analysis, solving multiple issues in parallel using Git worktrees. |
| Remote‑instance agents | Often hooked up for code‑review pipelines (Claude Code, Copilot, etc.). |
| “Pure vibe coding” | An extreme approach that many warn against due to the high risk of bugs in production. |
2. Decision‑Making Criteria
When I evaluate a new tool, I look at the following aspects:
| Aspect | What to Consider | Example / Notes |
|---|---|---|
| LLM model | • Latest, most capable model • Context‑window size • Cost per token | • Claude Sonnet 3.7 → favourite (mid‑2024) • Claude Sonnet 4 (same price, better) • Claude Opus 4 – great for large codebases but pricey • Mid‑July 2025: Grok 4 is “the best on the block” |
| Model support & diversity | Does the tool support multiple models or lock you into one? | • Cursor – supports many top models • Claude Code, Gemini CLI – built primarily for a single model |
| Editor / IDE integration | • VS Code vs. JetBrains vs. terminal • IDE‑specific features (e.g., JetBrains AI) | • Cursor – VS Code fork, tight editor integration • Claude Code – decent JetBrains IDE support • Codex & Gemini CLI – terminal‑centric |
| Speed / latency | Faster feedback → smoother dev experience | • JetBrains AI is the most integrated but one of the slowest • Slower tools increase feedback cycles, hurting productivity |
| Cost | • Pricing model (per‑use, token‑based, flat‑rate) • Unlimited vs. rate‑limited plans • Expected usage (light vs. heavy) | • Cursor: $20/mo (unlimited) → $60/mo Pro+ (daily users) → $200/mo Ultra (power users) • Claude Code: $100/mo Max plan (sweet spot for heavy users) • Gemini CLI: generous free tier (≈ $620/day) |
| Licensing & IP protection | • Company‑wide vs. individual licenses • Data‑training policies • Indemnity against IP claims | • Prefer company licensing (protects the organization) • Avoid tools that train on your data for commercial projects • Anthropic (Claude Code) indemnifies commercial users; Cursor only for >250 seats (as of writing) |
3. Model Landscape (mid‑July 2025)
| Model | Strengths | Weaknesses | Typical Use‑Case |
|---|---|---|---|
| Claude Sonnet 3.7 / 4 | Well‑tuned for software dev, decent context window | Smaller context vs. newer models | General coding assistance |
| Claude Opus 4 | Huge context, strong reasoning | Expensive per token | Large codebases, complex refactors |
| Grok 4 | Currently top‑performing, good coding insights | Newer, ecosystem still maturing | Cutting‑edge productivity |
| Gemini Pro (Google) | Massive free tier, integrates with Google Cloud | Trains on your data (free tier) | Hobbyist / open‑source projects |
| OpenAI GPT‑4‑Turbo | Broad ecosystem, many integrations | Cost can add up for heavy usage | General purpose, flexible |
4. Cost Overview (as of June 2025)
| Tool | Pricing Tier | What You Get | Typical User |
|---|---|---|---|
| Cursor | $20/mo (Pro) – unlimited (light) | VS Code fork, multi‑model support | Light users |
| $60/mo (Pro+) – higher limits | Faster usage, more tokens | Daily users | |
| $200/mo (Ultra) – “unlimited” (heavy) | Intended for power users, heavy Claude Opus 4 usage | Power users | |
| Claude Code | $100/mo (Max) – “unlimited” | Single‑model (Claude) with higher limits | Heavy users |
| Gemini CLI | Free tier – up to ~$620/day | Access to Gemini models, data‑training opt‑in | FOSS / hobbyists |
| GitHub Copilot | $10/mo (individual) | Inline completions, chat | General devs |
| JetBrains AI | Bundled with JetBrains IDEs (varies) | Deep IDE integration, slower response | JetBrains‑centric teams |
Note: “Unlimited” plans are usually rate‑limited; generous limits keep the experience smooth, but heavy usage of premium models (e.g., Claude Opus 4) can quickly exhaust quotas.
5. Licensing & IP Considerations
- Prefer company‑wide licensing – protects the organization rather than just the individual.
- Avoid tools that train on your data for commercial products.
- Example: Google Gemini CLI’s free tier does train on your code in exchange for a generous quota.
- Indemnity against IP claims – essential for enterprises.
- Anthropic (Claude Code) provides indemnification for commercial users.
- Cursor’s MSA only offers indemnity for customers with >250 seats (as of this writing).
Bottom line: If you work for a services firm or create client‑facing IP, steer clear of individual licenses and data‑training models unless you have explicit contractual protection.
Takeaways
- Tool choice is multidimensional – model quality, integration, speed, cost, and licensing all matter.
- Stay flexible – the “best” tool today (e.g., Grok 4) may be overtaken in months.
- Prioritise legal safety – especially for commercial or client‑driven work.
- Watch the price‑to‑performance ratio – heavy usage of premium models can make “unlimited” plans feel limited.
By mapping your needs against the criteria above, you can cut through the noise and pick a coding assistant that truly boosts your productivity without unexpected legal or financial surprises.
Choosing the Right Coding Assistant
For team members who are new to using coding assistants, start with GitHub Copilot. Its fixed‑cost model makes budgeting simple while you learn the fundamentals of:
- Prompt engineering
- Context engineering (more on these skills in another blog)
Once you’ve mastered these basics, consider moving to an API‑based tool that lets you switch between models.
- I’m a fan of Claude Sonnet and Claude Opus over OpenAI (and, to some extent, Gemini).
- If you can manage costs well, try Claude Code or an open‑source option such as OpenCode or Aider.
I would place OpenCode above Claude Code because of its flexibility.
If you need further clarification, reaching out to the tool’s support team is the best way to get help.
| Feature | Copilot | Claude | Open‑source | GitHub‑model | Other |
|----------------------------|------------------------|----------------------|---------------------------|-----------------------|-------|
| **Model support** | Multiple (incl. Grok 4) | Claude‑only | Multiple (open‑source) | GitHub‑model | Varies |
| **IDE integration** | VS Code fork | JetBrains IDEs | Terminal | VS Code/JetBrains | Varies |
| **Speed** | Fast | Moderate | Fast | Fast | Mixed |
| **Cost (typical heavy use)**| $60–$200/mo | $100/mo | Free / low‑cost | $10/mo | Varies |
| **Licensing** | Company‑wide options | Company‑wide (recommended) | Open‑source | Individual/Company | Varies |
| **Data training** | No (by default) | No (opt‑out) | No | No | Depends |