There's an exhausting amount of AI tool noise right now. Every week there's a new "game-changing" assistant, a new IDE plugin, a new "10x developer" claim. I've been filtering this for about two years now, and honestly most of it doesn't survive contact with real work.
This is my actual stack as of March 2026 — five tools I pay for (or use regularly on free tiers) and what I've genuinely learned about each of them. I'll keep it honest: these tools have real limitations, and I'll tell you what they are.
1. Claude Code (Terminal, $20/mo)
Claude Code is Anthropic's terminal-based coding agent. You install it via npm, point it at a repo, and it can read files, run commands, edit code, and execute multi-step tasks without needing a GUI.
What actually works: The orchestration mode is the standout feature. You can describe a refactoring task and it'll plan it, execute across multiple files, run tests, and fix failures — autonomously. For the kind of cross-cutting refactors that would take me half a day to do carefully, it completes in 15-20 minutes. The skills system (where it learns platform-specific patterns from previous sessions) is also genuinely useful for repetitive automation work.
What frustrates me: Context window management is on you. If the repo is large, you need to be deliberate about what files you include or it'll burn tokens fast and still miss relevant context. It also sometimes confidently makes changes that compile but break runtime behavior — you still need to review diffs carefully.
Who it's for: Developers who work on complex, multi-file tasks and are comfortable in the terminal. Not a replacement for understanding your codebase.
2. Gamma (Presentations, $10/mo)
I know, I know — a presentation tool in a dev tools list. But hear me out.
I regularly have to produce decks for non-technical stakeholders: architecture proposals, sprint retrospectives, feature pitches. Before Gamma, this meant either spending 2+ hours in Slides or producing something that looked like a 2008 PowerPoint.
What actually works: You describe what you want, paste in bullet points or a document, and Gamma generates a structured, reasonably designed slide deck in under a minute. The templates are clean and modern. For internal decks where design doesn't need to be pixel-perfect, it's a legitimate 2-hour time save.
What frustrates me: The AI-generated text tends toward corporate filler. It'll expand your bullet "reduce latency" into three sentences of generic performance narrative. You'll spend time pruning more than adding. Also, the free tier watermark is prominent — if you're pitching externally, you need the paid plan.
Who it's for: Developers who regularly communicate with non-technical audiences and want to stop losing half a day to slide design.
3. Kilo Code (VS Code Extension, Free–$20/mo)
Kilo Code is a VS Code extension that gives you access to 500+ models through a single interface with zero markup on API costs — you bring your own API keys (OpenAI, Anthropic, Google, local Ollama, whatever).
What actually works: The BYOM (bring your own model) approach makes it significantly cheaper than Cursor for high-volume usage. If you're using Claude Sonnet or GPT-4o heavily throughout the day, paying API rates directly instead of through a platform's markup adds up. It also supports local models through Ollama, which matters if you're working on anything sensitive that shouldn't leave your machine.
The multi-model switching mid-task is actually useful: start planning with a reasoning model, switch to a faster model for the actual code edits.
What frustrates me: Setup is more involved than Cursor. You need to manage API keys, understand which model is good for which task, and configure the context window settings yourself. Cursor is more plug-and-play. Kilo Code rewards developers who want control and are willing to invest setup time.
Who it's for: High-volume users or anyone who wants to use local models. Cursor is the better default recommendation for most people.
4. Notion AI (Workspace, $10/mo add-on)
I've used Notion for years as a team knowledge base, but the AI add-on only became worth it relatively recently.
What actually works: The workspace Q&A is the standout feature. You can ask "what's our current approach to rate limiting?" or "what did we decide about the auth architecture in Q4?" and it searches across all your Notion pages to answer. For teams with a lot of accumulated documentation, this replaces 10 minutes of searching with a 30-second question.
The "improve writing" and summarize features are fine but not exceptional — you can get the same quality from a chat interface. The value is in the integration with your existing workspace.
What frustrates me: At $10/month per user, it's expensive at team scale. A 5-person team is paying $600/year just for AI features on top of Notion's base plan. The ROI math only works if your team actually uses it regularly, and adoption is uneven in practice.
Who it's for: Teams that already run their knowledge base in Notion and have enough accumulated documentation that search is a real friction point.
5. Descript (Content Creation, $12/mo)
This one is niche — it's only relevant if you record screencasts, tutorials, or video documentation. But if you do, it's genuinely changed how I work.
What actually works: Transcript-based editing. Descript transcribes your recording and lets you edit the video by editing the text — delete a sentence from the transcript and the corresponding video section is removed. Filler word removal (ums, ahs, long pauses) is one click. For someone who records tutorial content or internal walkthroughs, this cuts editing time from hours to minutes.
The AI voice overdub feature is useful for fixing small verbal mistakes without re-recording — you type the correction and it generates audio in your voice.
What frustrates me: The overdub voice quality is good but not perfect — it works well for fixing a word or short phrase, but longer generated sections have a slightly synthetic quality that's noticeable. Also, the application is resource-heavy. On a machine with less than 16GB RAM, it can get sluggish on longer recordings.
Who it's for: Anyone who produces recorded technical content — tutorials, demo videos, documentation walkthroughs, or internal engineering knowledge sharing.
The Bottom Line
Here's what I'd recommend depending on your situation:
Solo developer doing heavy coding: Claude Code ($20/mo) + Kilo Code (API cost only). Skip the others until you have a team or content creation need.
Team lead or tech lead with stakeholder communication: Add Gamma ($10/mo) and Notion AI ($10/mo per user). The time saved on decks and document search compounds quickly.
Developer creating tutorial or educational content: Descript ($12/mo) is the highest-leverage tool for recorded content. The editing workflow change is real.
The honest version: AI tools save the most time on tasks that are repetitive, well-defined, or format-heavy. They still struggle with novel problem-solving and anything requiring deep understanding of your specific system. Use them to eliminate grunt work, not to replace thinking.
What's in your dev stack right now? Curious what people have found actually worth paying for.
Top comments (0)