Buyer guide · agentic harnesses
Coding-assistant myths in 2026: what the product pages blur
A myth-versus-reality check for coding assistants whose pricing, editor surfaces, privacy paths, and autonomy are easy to collapse into one misleading label.
Who this is for: Developers and buyers who have heard conflicting claims about Claude Code access, Cursor Hobby limits, Devin surfaces, OpenCode privacy, Continue’s roadmap, Copilot scope, agent review, MCP safety, and “free” OSS agents.
What actually matters
- Claude Code is not Max-only — Pro, Team, and Enterprise paths exist; the API is a separate billing route from chat subscriptions.
- Cursor Hobby is a limited entry plan, not an unlimited daily agent budget for teams.
- Devin Desktop (local IDE, formerly Windsurf) and Devin Cloud (browser delegated agent) are different products with different setup boundaries.
- OpenCode privacy depends on which connected provider, endpoint, and retention settings you choose — there is no single answer.
- Continue’s OSS extension remains useful for local/BYOK, but its hosted roadmap is frozen; new rollouts should plan alternatives.
- Copilot spans autocomplete, IDE Agent Mode, and a cloud coding agent — Neovim plugin remains completions-only.
- MCP servers grant real tool access; ignore files do not block them.
ACE recommendation
Use the exact product surface, account or API path, endpoint, and review gate as the source of truth. Pilot on a disposable branch, check live terms before spend or sensitive code, and treat autonomy as a reason for stronger review rather than less.
Set up the shortlist
How to run this decision
- 01Before purchase, open the exact product surface you will use (IDE plugin, CLI, cloud agent, API) and read live terms for that path.
- 02Build a small matrix: product × account type × data retention × training opt-out × included agent usage.
- 03Pilot on a disposable branch with non-sensitive code; verify billing and privacy on the actual account tier.
- 04Treat every agent patch as draft until a human reviews plan, tool calls, diff, tests, and rollback.
- 05Allowlist MCP servers before team rollout; review folder and network scopes per server.
- 06Re-check myths quarterly — plan names, quotas, and feature boundaries change with vendor releases.
Common pitfalls
- Buying based on a demo of one surface while the team will use a different billing path or client.
- Assuming “open source client” means zero cost — BYOK tokens, retries, and review time still add up.
- Skipping review because the marketing page says “autonomous developer.”
- Installing MCP servers without treating them like shell access grants.
- Migrating to Continue’s hosted path for new teams without reading the frozen-roadmap status.
Ask before you buy
- Myth: Claude Code is Max-only. Reality: Claude Pro, Team, and Enterprise can include Claude Code usage; the API is a separate billing path, while free Claude.ai does not include Claude Code.
- Myth: Cursor Hobby is a free unlimited agent. Reality: Hobby is a limited entry plan for Agent and Tab, so check current included usage and overage rules before treating it as a daily team budget.
- Myth: Devin Desktop and Devin Cloud are the same product. Reality: Desktop is the local IDE surface formerly known as Windsurf; Cloud is the browser-based delegated agent, with different setup and workflow boundaries.
- Myth: OpenCode has one privacy answer. Reality: build a small matrix for BYOK, OpenCode Go/Zen, and a local endpoint; the connected provider, retention terms, training settings, and data residency determine the actual boundary.
- Myth: Continue is still a growing hosted default. Reality: its OSS extension can remain useful for local or BYOK assistance, but the hosted roadmap is frozen/wound down, so new rollouts should evaluate migration alternatives.
- Myth: Copilot is only autocomplete. Reality: Copilot also has IDE Agent Mode and a cloud coding agent, but editor parity is uneven and the official Neovim plugin remains completions-only.
- Myth: an agent-generated patch is ready to merge. Reality: review the plan, tool calls, permissions, diff, tests, and rollback path before accepting any agent result.
- Myth: MCP servers are safe by default. Reality: treat each MCP install like granting shell and filesystem access — allowlist servers, review scopes, and log tool calls before team rollout.
- Myth: a free OSS agent is free to run daily. Reality: BYOK and local paths still bill tokens, retries, and operator review time; hosted passes are quota-based, not unlimited.
More buyer guides
Related decisions
Other harness buyer guides that answer the next question after this one.
Buyer guide
Editor vs terminal: which coding assistant workflow fits?
Choose the surface your team will actually use before comparing model quality.
Buyer guide
How to roll out an agentic harness without losing review discipline
A staged rollout pattern for teams that want speed without turning generated patches into unowned production risk.
Buyer guide
Agentic harness pricing: seats, tokens, and overage
Compare subscriptions and usage-based plans by the work they enable, not the number printed on the landing page.
Buyer guide
Enterprise AI coding assistant procurement checklist
A compact procurement block for security, legal, IT, and finance teams evaluating coding assistants beyond the demo and the seat price.
Buyer guide
Local LLM coding setup: Ollama, LM Studio, llama.cpp, and MLX
Pick a local inference stack and wire it to an agent harness without the usual traps — wrong ports, tiny default context, bad quants, and mismatched API prefixes that make agents look broken when the model is fine.
Buyer guide
Migrating from Roo Code to Kilo Code
Roo Code’s discontinuation pushed many teams toward Kilo’s OpenCode-lineage stack. This guide covers Marketplace install, credential migration, project rules, and the VS Code vs JetBrains paths without assuming settings auto-port.