Amp vs TabnineUpdated 2026-03-16

SUMMARY

Amp (CLI) uses markdown configuration while Tabnine (IDE) uses markdown. They differ on 4 of 11 compared features. Both tools provide AI-assisted development but take different approaches to project configuration.

FEATURE COMPARISON

FeatureAmpTabnine
Config formatmarkdownmarkdown
Hierarchy support
Global config
Project config
Subdirectory scoping
File inclusion / imports
Ignore file
IDE integrationCLI
Schema / structureFree-form (Markdown)Free-form (Markdown)
Git committed
EncodingUTF-8UTF-8

FILE MAPPING

PurposeAmpTabnine
Project instructionsAGENTS.md.tabnine/guidelines/*.md
Settings / configurationsettings.json.tabnine/agent/commands/

SIDE-BY-SIDE CODE SAMPLES

Amp
AGENTS.md
markdown
# Agent Instructions

## Project Context
Go microservices monorepo using Bazel for builds.

## Build & Test
- Build all: bazel build //...
- Test all: bazel test //...
- Integration tests: docker compose up -d && bazel test //integration/...
- Lint: golangci-lint run ./...

## Code Conventions
- All services follow standard layout in /cmd and /internal
- Use structured logging with slog
- Error wrapping with fmt.Errorf("op: %w", err)
- See @docs/api-style-guide.md for API conventions

## Cross-Repo References
- Shared proto definitions in sourcegraph/proto
- Common Go libraries in sourcegraph/lib

## Security
- Never log sensitive data (tokens, passwords, PII)
- All external inputs must be validated
- Use parameterized queries for database access

Tabnine
.tabnine/guidelines/*.md
markdown
# Application Guidelines

## Architecture
- This is a monorepo managed with Turborepo
- Apps: web (Next.js), api (Express), mobile (React Native)
- Shared packages: ui, config, types

## Code Standards
- TypeScript strict mode in all packages
- Use barrel exports (index.ts) for public APIs
- No default exports; use named exports everywhere
- Prefer async/await over raw Promises

## API Design
- RESTful endpoints follow /api/v1/{resource} pattern
- All responses use { data, error, meta } envelope
- Validate request bodies with Zod schemas
- Return 4xx for client errors, 5xx for server errors

## Database
- Use Prisma ORM for all database operations
- Migrations checked into source control
- Soft-delete pattern: use deletedAt timestamp, never hard delete
- All timestamps stored as UTC

## Testing
- Jest for unit tests, Playwright for E2E
- Test files colocated: foo.ts -> foo.test.ts
- Mock external services, never call real APIs in tests

KEY DIFFERENCES

1. Hierarchy support

Amp: Supported. Tabnine: Not supported.

2. Global config

Amp: Supported. Tabnine: Not supported.

3. Subdirectory scoping

Amp: Supported. Tabnine: Not supported.

4. IDE integration

Amp: CLI. Tabnine: Supported.

WHICH SHOULD I USE?

Recommendation

Choose based on your IDE/CLI preference. If you use Amp, configure AGENTS.md. If you use Tabnine, configure .tabnine/guidelines/*.md. Many projects include config files for multiple tools so each team member can use their preferred editor.

PORTABILITY TIP

sync-instructions.sh
bash
# Maintain a single source of truth:
cp AI-INSTRUCTIONS.md CLAUDE.md
cp AI-INSTRUCTIONS.md .cursorrules
cp AI-INSTRUCTIONS.md AGENTS.md

SEE ALSO

Tool pages:
File references:
agentconfig.ing2026-03-16COMPARE(1)