AI Tools to Start With: The Non-Hype Stack for Actually Getting Things Done

The AI tool landscape is overwhelming on purpose. There are thousands of tools, most of them built on the same foundation models, most of them good enough to be useful and not good enough to be transformative.

This guide cuts through. These are the tools that actually produce leverage — based on what's working for people building real things right now, not what's trending on Twitter.

Orientation: What You're Actually Choosing

Every AI tool wraps around one of a few foundation models (GPT-4o, Claude, Gemini, Llama). The wrapper you choose matters for:

• Workflow integration: Does it fit naturally into how you work?
• Context handling: How much it can hold in memory at once
• Output formatting: Does it produce structured output or conversational text?
• Cost: API costs, subscription models, and usage limits

Most AI productivity problems aren't "wrong tool" problems — they're "poor prompting" problems. Learning to direct foundation models effectively is worth more than collecting specialized tools.

That said, here's the honest stack.

Foundation Layer: The Models You Should Know

Claude (Anthropic) — The best model for long-form writing, nuanced reasoning, and following complex instructions. Claude's context window (200K tokens) means it can hold and analyze entire documents, codebases, or research bodies in a single session. Best for: writing, research synthesis, complex multi-step reasoning, anything where you need the AI to "understand" a large amount of context.

GPT-4o (OpenAI) — The most capable generalist model with the broadest ecosystem. Best for: integration with other tools (via OpenAI's API), image analysis, voice, and anything requiring the widest capability surface. Also has the most mature plugin and tools ecosystem.

Gemini 1.5 Pro (Google) — The leader in multimodal work and the longest context window (1M tokens). Best for: analyzing video, very large documents, and workflows that integrate with Google Workspace.

Llama 3 (Meta, open source) — The best option for self-hosting, privacy-sensitive applications, or building customized fine-tuned models without API costs.

Recommendation: Start with Claude for writing and research, GPT-4o for general use and integration. Add Gemini for multimodal and large-context work.

Writing and Content

Claude via Claude.ai (or API) — The default choice for any serious writing work. The quality at long-form reasoning and nuanced writing is unmatched. Use it for research synthesis, first drafts, editing feedback, and structural planning.

Perplexity — The best AI search engine. Unlike ChatGPT's training cutoff, Perplexity searches the real-time web and cites sources. Invaluable for research that requires current information. Use it instead of Google for initial research on most topics.

Lex — A writing tool with AI embedded in the workflow itself. Less chat-interface, more collaborative writing partner in a document. Good for writers who find the chat interface breaks their writing flow.

Coding and Building

Cursor — An IDE (code editor) with AI deeply embedded. Claude and GPT-4 are integrated directly into the editing experience — not as a sidebar, but as context-aware assistance that understands your entire codebase. If you're building software, Cursor is the highest-leverage coding tool available right now.

Claude via API — For serious programming work, using Claude directly via API (in Cursor or via scripts) gives the most control over context and prompting. The ability to put your entire codebase context into a 200K-token window is transformative for complex debugging and refactoring.

Bolt / v0 — For rapid prototyping of web interfaces without deep coding experience. These tools turn text descriptions into functional frontend code. Best for: quickly testing UI concepts, building simple tools, or generating a starting point for a developer to refine.

Research and Analysis

Perplexity Pro — Real-time web search with AI synthesis. Essential for any research that needs current data. The Pro tier lets you choose which foundation model handles the synthesis.

NotebookLM (Google) — Upload documents, research papers, or your own writing, and create an AI that's specifically trained on those sources. It generates summaries, answers questions about the documents, and even creates audio podcast-style overviews. Invaluable for synthesizing large bodies of research.

Elicit — AI designed specifically for academic research. Searches academic papers, extracts key findings, and synthesizes across studies. If you need to understand what the research says on a specific scientific or academic topic, Elicit is significantly better than asking ChatGPT.

Image and Visual

Midjourney — Still the quality leader for photorealistic and artistic image generation. The Discord-based interface is clunky, but the output quality for creative work is unmatched. Best for: brand imagery, conceptual visualization, creative assets.

DALL-E 3 (via ChatGPT) — The most integrated and accessible image generation option. Built into ChatGPT, understands nuanced text descriptions, and handles specific composition requests better than competitors. Best for: quick generation, concept exploration, when you need image + chat in one workflow.

Adobe Firefly — The safest choice for commercial work. Trained on licensed images, so there are no copyright exposure issues. Integrated into Adobe's tools. Best for: professional creative work where IP clarity matters.

Productivity and Workflow

Notion AI — If you use Notion (a significant share of knowledge workers), the AI integration is genuinely useful. Drafting, summarizing, extracting action items from meeting notes — it's embedded where the work already lives.

Otter.ai / Fireflies — AI meeting transcription with summaries and action item extraction. Eliminates the "taking notes during the meeting" problem. If you're in a lot of calls, this is one of the fastest-payback AI tools.

Make (formerly Integromat) — AI-enhanced workflow automation. More powerful than Zapier for complex workflows, and the AI integration layer lets you add AI processing steps into your automation chains (summarize a document, classify an email, generate a draft response).

The Honest Caveat

Every tool on this list will be partially obsolete in 12–18 months. The foundation models are improving rapidly. The tooling ecosystem is consolidating. Things that require specialized tools today will be built into foundation model interfaces tomorrow.

The durable skill isn't mastering specific tools. It's developing the judgment to evaluate AI output quality, the prompting fluency to direct AI systems effectively, and the workflow architecture skill to build AI into your process.

Tools change. Judgment compounds.

Where to Start If You're New

If you're just starting: Claude.ai and Perplexity cover 80% of the use cases that produce real leverage for most knowledge workers. Get excellent at using both before adding complexity.

The goal isn't to use more AI tools. It's to do better work. Start with what produces real results and expand from there.

Key Takeaways

• Most AI productivity problems are prompting problems, not tool-selection problems — learn to direct foundation models before optimizing your tool stack
• Foundation layer recommendation: Claude (writing, research, reasoning), GPT-4o (general use, integration), Gemini (multimodal, large context)
• Key stack: Claude for writing, Perplexity for research, Cursor for coding, NotebookLM for document synthesis, Midjourney or DALL-E 3 for images
• The durable skill is AI fluency and workflow judgment — specific tools become obsolete faster than judgment does
• Start simple: Claude.ai + Perplexity covers 80% of knowledge worker leverage before adding complexity

---

Part of the Abundance OS framework — the definitive guide to exponential AI, energy, and the collapse of scarcity.