The Stack

Nine tools. Nine different jobs. We know which is which.

Every AI tool markets itself as the one you need. In practice, each one is genuinely better at a narrower slice of the work — this is the honest breakdown we use ourselves to decide what runs your project.

Field notes based on how we actually use these tools day to day. Pricing and capabilities move fast — treat this as directional, not a spec sheet frozen in time.

Claude Code01
Anthropic — terminal agent
Best for
Full site & app builds, multi-file refactors, autonomous multi-step work
Not built for
Quick inline edits while you type
Our use
Primary build agent for larger sites and apps
Claude02
Anthropic — chat & writing
Best for
Long-form copy, strategy documents, natural-sounding brand voice
Not built for
Real-time web-current facts without search enabled
Our use
Content, positioning, and client-facing copy
Codex03
OpenAI — terminal agent
Best for
Fast scripted tasks, lean background jobs, ChatGPT-ecosystem workflows
Not built for
Holding an entire large codebase in context at once
Our use
Quick builds and fallback when another agent is rate-limited
GPT04
OpenAI — general purpose
Best for
General productivity, creative drafts, one tool that does a bit of everything
Not built for
Deep autonomous coding without its agent tooling
Our use
Research drafts and general client communication
Gemini05
Google — reasoning & multimodal
Best for
Deep research, large-context analysis, native Google Workspace work
Not built for
Polished document/slide output out of the box
Our use
Research-heavy briefs and data-dense client work
Copilot06
Microsoft/GitHub — inline assistant
Best for
In-editor autocomplete, teams standardized on GitHub or Microsoft 365
Not built for
Deep autonomous multi-file work on its own
Our use
Client handoff when their team already lives in that ecosystem
Qwen07
Alibaba — open-weight
Best for
Self-hosted or cost-sensitive workloads, custom fine-tuned deployments
Not built for
Turnkey polish without extra engineering around it
Our use
Budget-constrained automation where we control the hosting
Replit08
Replit — build & deploy
Best for
Fast prototypes with instant hosting, non-coder-friendly app builds
Not built for
Heavy production infrastructure at scale
Our use
Rapid client prototypes we can demo the same day
Manus09
Autonomous agent
Best for
End-to-end tasks that need browsing, files, and code with no hand-holding
Not built for
Ambiguous, open-ended briefs without a clear task shape
Our use
Research and data-collection tasks run in the background

If you only remember one line per tool

Building a full site or appClaude Code
Writing brand copy or strategyClaude
Quick script, fast turnaroundCodex
General research & draftsGPT
Deep research, big contextGemini
Team already on GitHub/365Copilot
Self-hosted, cost-sensitiveQwen
Prototype fast, demo todayReplit
Hand off a whole task, walk awayManus