Note: This page is under construction. I am actively revising the recommendations, structure, and personal comments.

This page functions as my current reference for AI tools. Make sure that you confirm the date of this article to check for relevance. The space can change a lot in a short amount of time.

Disclosure: I used AI to go out onto the internet and do some research to find new tools to review. Personal comments are all written by me.

Practical Shortlists

Day to Day Tools

  • General Assistant: GPT models, Claude, Gemma-4.
  • Meeting Notes: Whisper and its associated alternatives.
  • Writing: Claude, Harper, ChatGPT.
  • Slides And Design: Use any LLM to write to Marp format.
  • Search And Research: Perplexity.

Developer Specific

  • IDE Assistant: Cursor.
  • Terminal Or Code Agent: Pi, Codex, Claude Code.
  • Local Models: Gemma-4, Qwen3.6
  • Inference Serving: vLLM, llama.cpp.
  • RAG Apps: LangChain / LangGraph.
  • Vector Database: Pinecone.

Tools For Organizations

  • Enterprise Chat: Frontier labs have enterprise plans.
  • Secure App Building: AWS Bedrock, Anthropic API.
  • Self-Hosted Or Private Workflows: LiteLLM, vLLM.

Decision Trees

  • One general AI subscription? Start with ChatGPT. Compare Claude if you write long documents or code often.
  • Are you mostly writing or reviewing long-form text? Try Claude first.
  • Are you doing sourced public research? Start with Perplexity, then verify sources directly.
  • Are you coding inside an editor? Start with Cursor.
  • Are you delegating repo work to a terminal agent? Try Codex or Claude Code.
    • As of this article, Codex is likely better to use day to day. Claude is easier to onboard with.
  • Do you want an extensible, minimal, open-source, provider-independent harness? Try Pi.
  • Do you want a visual interface for many agent threads? Try T3 Code.
  • Does the work touch sensitive data? Use enterprise-licensed tools with contractual controls, or host your own models.
  • Does it need access to internal files, email, calendar, or code? Claude with enterprise tools.

Buying Advice

  • Pay for one general assistant first. Just pick one service to start.
  • Add a coding tool only if you code every week. Cursor or Claude is easier to justify if you code all the time.
  • Use Perplexity when sources matter. It is the strongest research companion.
  • Prefer reviewable workflows. The best tools can integrate with Git, and you can review their output.
  • Watch usage-based pricing. Bring-your-own-key and agentic tools can burn money faster than autocomplete tools.
  • Review subscriptions monthly. AI tools overlap heavily, and it is easy to pay for three tools when one would do.

General Assistants

If you are new to the space, I would start with either Codex/GPT or the Claude models. Additionally, other services are outlined below.

  • ChatGPT: Best default if you want one general AI subscription for writing, brainstorming, coding help, images, documents, and everyday questions. It is broad, polished, and easy to recommend to someone who does not want to think too hard about the category.
    • Best For: General-purpose AI help; first drafts and rewrites; explanations; mixed text, code, image, and document workflows.
    • Caveats: It can sound confident when it is wrong. Ask for sources when facts matter, and compare it with Claude if your main work is code or long documents.
    • Personal Comments: TODO
  • Claude: Best when reasoning quality, prose quality, long context, or careful code review matters more than having the broadest product ecosystem. MAKE SURE TO MENTION CLAUDE CODE
    • Best For: Coding help; code review; long documents; careful editing; refactoring plans; trade-off analysis.
    • Caveats: It is still not a substitute for tests, review, or domain judgment. Usage limits can matter in long sessions.
    • Personal Comments: TODO
  • Gemini: Can be useful when doing multimodal work like summarizing YouTube videos.
    • Personal Comments: Not my favourite. I pretty much avoid it as it’s strictly worse than other models.
  • Perplexity: Best for doing a “shotgun approach” to get your literature review off the ground.
    • Caveats: Can still hallucinate. Be careful.
    • Personal Comments: I have only used this a few times. It worked well for AI research. Your mileage may vary.
  • Grok, Microsoft Copilot, Mistral Le Chat, DeepSeek, Qwen, Kimi, GLM, And Other Frontier Models: These are worth keeping an eye on to see how they are performing relative to the latest models.
    • Best For: Model comparison, privacy, open-weight experimentation.
    • Personal Comments: Personally, I have found uses for Gemma in my day to day work. Copilot on the other hand I have little to no use for.

Coding Tools

There’s a few different major categories of coding tools at the moment. Some exist in multiple categories. Pick what seems appealing and don’t be afraid to change your mind later.

  • Editor based: Extensions to things like VS Code. Most of the frontier labs support this.
  • Terminal: Some sort of TUI application outside of an editor.
  • Open-source: Harnesses that are open
  • Cloud agents: Stay away unless you have a good reason. Can really burn tokens.

Editor-First Tools

  • Cursor: AI-native editor with chat, multi-file edits, codebase search, and agentic workflows in one polished interface.
    • Best For: Developers who want an AI-native editor; multi-file changes; fast prototyping; people willing to move into a new editor.
    • Caveats: Watch pricing and usage limits. It can make large changes quickly, so commit often and review diffs carefully.
    • Personal Comments: TODO
  • GitHub Copilot: Practical default if you want AI inside your existing editor with minimal workflow change.
    • Best For: Keeping your current editor; autocomplete; small edits; teams already using GitHub.
    • Caveats: It is not always the strongest option for large autonomous refactors. Teams should review policy, licensing, privacy, and training settings.
    • Personal Comments: TODO
  • Windsurf: Cursor-like AI editor worth comparing on pricing, model access, and workflow feel.
    • Best For: AI-native editor workflows; Cursor alternatives; fast application prototyping; integrated editor and agent experiences.
    • Caveats: It overlaps heavily with Cursor, so the main question is fit, not whether the category is useful.
    • Personal Comments: TODO
  • Cline, Roo Code, And Continue: VS Code-based options for people who want more control over models, prompts, approval steps, or bring-your-own-key setups.
    • Best For: VS Code users; open-source or configurable workflows; careful approvals; testing multiple model providers.
    • Caveats: Configuration takes more work than Copilot or Cursor, and bring-your-own-key costs can be harder to predict.
    • Personal Comments: TODO

Terminal And Harness Tools

  • Claude Code: Terminal coding agent that can inspect a repo, edit files, run commands, and iterate with tests.
    • Best For: Real codebase maintenance; refactors with reviewable diffs; test-driven fixes; terminal-first developers.
    • Caveats: Keep changes small, run tests, and review every diff. Do not give broad agent access to secrets or production systems.
    • Personal Comments: TODO
  • OpenCode: Open-source terminal coding agent with broad model-provider support.
    • Best For: Terminal-first coding; bring-your-own-model workflows; local or alternative providers; configurable permissions.
    • Caveats: You are responsible for model choice, API keys, and cost control. Expect more setup than a bundled subscription.
    • Personal Comments: TODO
  • Aider: Mature Git-native pair-programming tool that keeps AI changes close to commits, diffs, and explicit file context.
    • Best For: Git-centered development; small and medium changes; terminal pair-programming; transparent diffs.
    • Caveats: It rewards users who understand Git well. Model quality and cost depend on what you connect to it.
    • Personal Comments: TODO
  • pi: Minimal terminal coding harness built around composability: tools, extensions, skills, prompt templates, themes, model providers, API usage, and SDK integration.
    • Best For: Custom coding-agent workflows; provider-agnostic model use; automation; scripting; RPC; SDK and package-based extension work.
    • Caveats: It is more primitives-first than turnkey. If you want a polished editor assistant, start with Cursor or Copilot.
    • Personal Comments: TODO
  • Codex CLI And Gemini CLI: Official terminal paths into OpenAI and Google coding workflows.
    • Best For: Trying model-specific coding workflows; lightweight terminal tasks; comparing model ecosystems; using existing credits or subscriptions.
    • Caveats: They may be less configurable than OpenCode, Aider, or pi, and value depends on current model quality and plan limits.
    • Personal Comments: TODO
  • Goose: Open-source agent broader than coding, with desktop, CLI, and workflow automation use cases.
    • Best For: Open-source agent experimentation; workflow automation beyond code edits; internal agent-platform exploration.
    • Caveats: It may be more platform than focused coding assistant. Permission boundaries matter.
    • Personal Comments: TODO

Agent Interfaces And Cloud Agents

  • T3 Code: Open-source desktop control surface for managing coding agents across repositories or parallel sessions.
    • Best For: Comparing agents side by side; multi-repository sessions; parallel agent workflows; visual control over CLI agents.
    • Caveats: It depends on the underlying agents and models you connect, and it adds another layer to debug.
    • Personal Comments: TODO
  • Devin, Replit Agent, And Cloud Agents: Managed remote environments for prototypes, isolated tasks, and delegated implementation work.
    • Best For: Greenfield prototypes; isolated tasks with clear acceptance criteria; browser-based development; remote agent execution.
    • Caveats: They need clear instructions and careful review. Be cautious with sensitive repositories, credentials, and production access.
    • Personal Comments: TODO

Local And Self-Hosted Inference

Use these when you want more privacy, local experimentation, cheaper high-volume inference, or control over model hosting. Most non-technical users can skip this category.

Single-User And Developer Tools

  • Ollama, LM Studio, Jan, GPT4All, And llama.cpp: These are the most relevant local-AI tools for individual users and developers. Ollama is the easiest terminal-first default, LM Studio is the polished GUI, Jan and GPT4All are privacy-friendly desktop options, and llama.cpp is the low-level engine underneath much of the local ecosystem.
    • Best For: Local model testing; private drafts; offline experimentation; comparing open-weight models; developer sandboxes.
    • Caveats: Local models still depend on your hardware. Smaller models can be useful, but they are not a drop-in replacement for the best hosted frontier models.
    • Personal Comments: TODO
  • LocalAI, text-generation-webui, KoboldCpp, Llamafile, Lemonade, node-llama-cpp, And Docker Model Runner: These are more specialized tools for people who already know why they want them.
    • Best For: Power-user web UIs; OpenAI-compatible local APIs; creative-writing setups; portable model bundles; AMD acceleration; JS integrations; containerized local models.
    • Caveats: Expect more setup, rougher edges, and more responsibility for model selection.
    • Personal Comments: TODO

Production Serving And Tuning

  • vLLM, SGLang, TensorRT-LLM, TGI, And LMDeploy: These are production-serving tools for teams running models at scale.
    • Best For: Multi-user inference; batching; throughput; OpenAI-compatible internal endpoints; optimized GPU serving.
    • Caveats: This is infrastructure work, not a consumer buying decision. You need operations skill and a real reason to self-host.
    • Personal Comments: TODO
  • Unsloth, Axolotl, And ExLlamaV2: These are optimization and fine-tuning tools for teams customizing or accelerating open models.
    • Best For: QLoRA fine-tuning; production training pipelines; GPTQ or EXL2 inference.
    • Caveats: Fine-tuning is easy to overestimate. Try prompting, retrieval, and workflow changes before training your own model.
    • Personal Comments: TODO

Research And Literature Tools

Use these when you need evidence, papers, citations, or source-grounded synthesis rather than general brainstorming.

  • Elicit, Consensus, Semantic Scholar, Scite, SciSpace, ResearchRabbit, Connected Papers, PapersFlow, And Zotero: These are the core academic-research tools I would compare. Elicit is good for structured extraction, Consensus is good for evidence questions, Semantic Scholar is a strong free index, Scite helps with citation context, SciSpace explains dense papers, ResearchRabbit and Connected Papers map related work, PapersFlow helps with literature-review drafting, and Zotero remains the reference-management anchor.
    • Best For: Literature reviews; paper discovery; citation trails; evidence comparison; research organization.
    • Caveats: Do not outsource judgment to any research tool. Read the actual papers when the claim matters.
    • Personal Comments: TODO
  • NotebookLM And Deep Research Tools: NotebookLM is useful when you want source-grounded synthesis from uploaded documents. Deep Research-style tools are useful for broader scoping.
    • Best For: Uploaded PDFs; source-grounded summaries; meeting notes; quick research briefs; starting bibliographies.
    • Caveats: Source-grounded does not mean perfect. Check quotes, citations, and omissions.
    • Personal Comments: TODO

Image, Video, And Audio Generation

This category changes fast. I would choose based on the kind of media you actually make, not on a single viral demo.

Image

  • Midjourney, FLUX, Stable Diffusion, Imagen, DALL-E, Adobe Firefly, Ideogram, Leonardo.ai, And Getty Generative AI: Midjourney is still known for aesthetic output, FLUX and Stable Diffusion matter for open and API-driven workflows, Imagen and DALL-E are convenient inside larger ecosystems, Firefly and Getty aim at commercial safety, Ideogram is strong for typography, and Leonardo.ai is creator-friendly.
    • Best For: Concept art; marketing images; social assets; product mockups; typography experiments; controllable open workflows.
    • Caveats: Rights, likeness, training data, and commercial-use rules matter. Read the license before using outputs in paid work.
    • Personal Comments: TODO

Video

  • Veo, Runway, Kling, Sora, Pika, Luma, Seedance, Hailuo, Wan, HeyGen, Synthesia, Hedra, D-ID, And Stable Video Diffusion: Video is split between cinematic generation, social-first clips, open-weight experimentation, and avatar or talking-head tools.
    • Best For: Short clips; ad concepts; storyboards; character references; avatar videos; product explainers; experimental filmmaking.
    • Caveats: Quality, pricing, rights, and availability shift constantly. Sora’s consumer app and API are marked as being discontinued in 2026 in the landscape notes, so do not build a long-term workflow around it without checking current status.
    • Personal Comments: TODO

Audio

  • ElevenLabs, Play.ht, Resemble AI, WellSaid Labs, Murf AI, Descript, Coqui TTS, Whisper, Voxtral, Suno, Udio, ElevenLabs Music, Stable Audio, AIVA, MusicGen, And Lyria: Voice tools are useful for narration, dubbing, transcription, and voice cloning. Music tools are useful for demos, backing tracks, and creative experiments.
    • Best For: Voiceovers; transcription; dubbing; podcast editing; sound design; song sketches; background music.
    • Caveats: Voice cloning, music rights, likeness rights, and platform rules matter. Be especially careful with commercial release.
    • Personal Comments: TODO

Agents And Automation Frameworks

These are for builders. If you are not building AI workflows or internal tools, you can skip most of this.

Frameworks And SDKs

  • LangGraph, CrewAI, AutoGen / AG2, Microsoft Agent Framework, Semantic Kernel, LlamaIndex, Pydantic AI, Smolagents, MetaGPT, OpenAgents, OpenAI Agents SDK, Claude Agent SDK, Google ADK, Vercel AI SDK, Mastra, And Mirascope: This is the builder landscape for agent orchestration. LangGraph is the production default I would investigate first, CrewAI is approachable for role-based prototypes, Microsoft Agent Framework is important for Microsoft shops, LlamaIndex is strongest around retrieval, and the vendor SDKs are useful when you are committed to one ecosystem.
    • Best For: Stateful agents; multi-agent workflows; retrieval-grounded agents; typed agent apps; vendor-native agents; production orchestration.
    • Caveats: Frameworks add complexity quickly. Start with a boring workflow and add orchestration only when the problem requires it.
    • Personal Comments: TODO

Computer Use, Workflow Automation, And Observability

  • Anthropic Computer Use, OpenAI Operator / ChatGPT Agent, Browser Use, Playwright MCP, n8n, Zapier, Make, Dify, Flowise, Langflow, LangSmith, Langfuse, Arize Phoenix, Helicone, And Braintrust: These cover browser control, no-code or low-code automation, visual LLM app building, and eval or tracing.
    • Best For: Browser tasks; workflow automation; internal prototypes; visual app builders; tracing; cost tracking; evaluation.
    • Caveats: Automation without observability becomes fragile fast. Anything that can act in a browser or workflow tool needs permission boundaries.
    • Personal Comments: TODO

RAG, Vector Databases, And AI Infrastructure

This category matters when you are building retrieval systems, internal search, document workflows, or production LLM apps. Most buyers should not start here.

  • Pinecone, Qdrant, Weaviate, Milvus, Chroma, pgvector, LanceDB, Zilliz Cloud, Deep Lake, Faiss, Vespa, Elasticsearch, OpenSearch, And Turbopuffer: These are vector search and retrieval storage options. Pinecone is managed and easy to adopt, Qdrant is a strong open-source default, pgvector is attractive if you already live in Postgres, Chroma is good for prototypes, and Milvus or Zilliz matter at larger scale.
    • Best For: Semantic search; RAG; recommendation systems; internal knowledge bases; hybrid retrieval; production vector storage.
    • Caveats: A vector database does not fix bad data. Chunking, metadata, permissions, evaluation, and freshness usually matter more than the logo.
    • Personal Comments: TODO
  • OpenAI Embeddings, Voyage AI, Cohere Embed, BGE, E5, nomic-embed-text, Jina Embeddings, LlamaIndex, LangChain, Haystack, DSPy, RAGFlow, Verba, Quivr, Khoj, Unstructured.io, LlamaParse, Firecrawl, Reducto, Docling, OpenRouter, LiteLLM, Portkey, And MCP Servers: These are the surrounding pieces for embeddings, retrieval frameworks, document ingestion, model routing, and tool integration.
    • Best For: Document parsing; retrieval pipelines; model routing; OpenAI-compatible gateways; local knowledge apps; MCP-based tool use.
    • Caveats: Each extra layer needs a reason. Keep the first system boring and measurable.
    • Personal Comments: TODO

Writing, Editing, And Documents

For workplace writing, the meaningful distinction is not only which tool writes best. It is where the data goes, whether prompts and outputs are retained, whether enterprise controls exist, and whether the tool fits the document system people already use.

  • ChatGPT And Claude: Strong general-purpose tools for drafting, rewriting, structure, long-form prose, and technical explanation.
    • Best For: Drafts; rewrites; outlines; technical explanations; long-form editing; policy and documentation work.
    • Caveats: Do not paste sensitive material into consumer plans unless policy allows it. Review important writing for accuracy, tone, and missing context.
    • Personal Comments: TODO
  • Grammarly, Notion AI, Google Gemini In Docs/Gmail, And Microsoft Copilot In Word/Outlook: Embedded writing tools that live where work already happens.
    • Best For: Email; documents; notes; workplace tone; grammar; summaries; enterprise document workflows.
    • Caveats: Integration is useful, but governance still matters. Check retention, training, admin controls, and sharing defaults.
    • Personal Comments: TODO
  • Lex, Sudowrite, Writer, Jasper, And Copy.ai: Specialized writing and marketing tools.
    • Best For: Creative writing; brand-controlled copy; marketing drafts; campaign variants.
    • Caveats: They can produce polished generic prose. Brand quality still needs human taste and review.
    • Personal Comments: TODO

Meetings, Transcription, And Productivity

Meeting bots are high-risk because they capture sensitive spoken content that people may not think of as documents. Treat them as data-capture systems, not just convenience tools.

  • Otter.ai, Fireflies.ai, Fathom, Granola, tl;dv, Zoom AI Companion, Microsoft Teams Copilot, And Google Meet Gemini: Meeting transcription, summaries, searchable meeting histories, and meeting-native assistant features.
    • Best For: Meeting notes; searchable transcripts; CRM handoffs; project follow-up; Zoom, Teams, or Google-native summaries.
    • Caveats: Consent, retention, storage, sharing, and admin controls matter. Some meetings should not have a bot in them.
    • Personal Comments: TODO

Data Analysis, BI, And Analytics

The serious limitation here is semantic correctness. AI over dirty BI models produces confident nonsense. These tools work best when definitions, permissions, and data models are already managed.

  • ChatGPT, Claude, Julius AI, Power BI Copilot, Tableau Agent, Databricks Genie, Hex, Deepnote, ThoughtSpot Spotter, Sigma, Qlik, And Looker Gemini: File analysis, spreadsheet help, notebook workflows, natural-language BI, and enterprise analytics assistants.
    • Best For: Ad hoc CSV analysis; spreadsheet explanation; chart drafts; BI Q&A; collaborative notebooks; lakehouse analytics.
    • Caveats: If the semantic layer is wrong, the AI answer will be wrong. Permissions, data lineage, and metric definitions matter.
    • Personal Comments: TODO

Enterprise Search And Knowledge Assistants

For organizations, this category can matter more than consumer chatbots because it touches internal knowledge, permissions, retention, and auditability.

  • Glean, Microsoft 365 Copilot, Google Gemini Enterprise, ChatGPT Enterprise / Team, Claude Enterprise / Team, Hebbia, Dust, Dify, Flowise, LangChain, LangGraph, LlamaIndex, Haystack, Pinecone, Weaviate, Qdrant, Chroma, Elastic, And OpenSearch: Enterprise search, document analysis, internal assistants, RAG frameworks, vector databases, and hybrid search systems.
    • Best For: Internal search; governed assistants; document-heavy analysis; legal and finance workflows; custom internal AI apps; permission-aware retrieval.
    • Caveats: Connecting to data is not enough. The system must enforce permissions, log usage, support retention rules, and avoid training on your content unless explicitly approved.
    • Personal Comments: TODO

Developer Platforms And APIs

These matter when you are building AI into products or internal systems rather than buying an end-user app.

  • OpenAI API, Anthropic API, Google Vertex AI / Gemini API, Azure OpenAI / Azure AI Foundry, AWS Bedrock, Mistral AI, Cohere, Together AI, Fireworks, Groq, Cerebras, Hugging Face, Replicate, Modal, Baseten, And BentoML: Model APIs, enterprise cloud AI platforms, hosted open-weight inference, model hubs, and deployment infrastructure.
    • Best For: Product features; internal tools; enterprise deployments; multimodal apps; hosted open models; custom model deployment.
    • Caveats: Vendor choice affects data handling, latency, cost, model availability, regional requirements, and operational complexity.
    • Personal Comments: TODO