The Problem
If you've tried to evaluate AI agent frameworks recently, you know the drill: every tool's landing page says it's production-ready, every benchmark shows it winning, and every tutorial uses a trivially simple example.
I hit this wall repeatedly as a Technical Director responsible for technology decisions. My evaluation process was scattered -- browser tabs, Notion pages, Slack threads, half-remembered conference talks. When a colleague asked "should we use LangGraph or CrewAI?", I was reconstructing my analysis from memory.
So I started building a structured system. It grew into something I think other developers would find useful, so I put it online.
What I Built
TekAI is a tech intelligence hub with three main components:
1. Catalog (101 entries and counting)
Every tool, framework, vendor, and pattern I've evaluated gets a structured review:
| Category | Entries |
|---|---|
| AI & Machine Learning | 65 |
| Infrastructure | 15 |
| Security | 7 |
| Platforms | 6 |
| DevOps | 4 |
| Auth | 2 |
| Databases | 2 |
Each entry includes: radar ring assessment, hands-on evaluation notes, competitive context (who are the alternatives?), relevant article links, and tags for discovery.
2. Interactive Tech Radar
Inspired by ThoughtWorks' technology radar, but focused on the tools developers actually touch daily. Four rings:
- Adopt (1 entry) -- Strong confidence, actively using in production
- Trial (19 entries) -- Worth investing serious evaluation time
- Assess (79 entries) -- Interesting but unproven or too narrow for broad recommendation
- Hold (2 entries) -- Proceed with caution, known issues
The visualization is an interactive SVG -- click any dot to jump to the full review. Color-coded by category, with angular distribution to minimize overlap.
3. Article Reviews (36 and growing)
I read vendor announcements, research papers, and technical deep-dives, then write structured reviews with credibility ratings:
- High credibility -- Original research, reproducible results, balanced analysis
- Medium credibility -- Useful information but with caveats (limited scope, potential bias)
- Low credibility -- Marketing dressed as analysis, unreproducible claims, or missing methodology
This saves time by letting you skip straight to the high-credibility sources.
4. Solution Finder
Describe what you're building -- AI agent, secure runtime, model evaluation pipeline -- and answer a few focused questions. The finder scores catalog entries against your requirements and explains why each recommendation fits.
The Methodology
The radar assessment is opinionated. That's intentional. Here's my framework:
Adopt means I've used it in production, understand the failure modes, and would recommend it to a peer without caveats.
Trial means I've done enough evaluation to believe it's worth your time, but I haven't seen it through a full production lifecycle yet.
Assess means it's technically interesting, solves a real problem, but has unknowns -- maturity, community, maintenance trajectory, or edge cases I haven't tested.
Hold means I've identified concrete concerns -- not that it's bad, but that you should be aware of specific risks before adopting.
The fact that 79 of 101 entries sit in Assess is a feature, not a bug. Most tools are genuinely too new or too narrow for a confident recommendation. Saying "I don't know yet" is more useful than a premature endorsement.
Tech Stack
The site itself is built with Astro v6 (static output), Tailwind CSS v4 for styling, and deploys as a static site. The radar visualization is pure JS -- no D3 dependency. Content is managed as markdown files with structured frontmatter.
What's Next
I update the catalog regularly as I review new tools and revisit earlier assessments. Radar rings change when my confidence changes -- and I document why.
Upcoming areas I'm expanding:
- More infrastructure and security coverage (currently underrepresented relative to AI/ML)
- Longitudinal trend tracking (which tools are gaining momentum vs stalling)
- Community-suggested reviews
Explore the full catalog and radar: tekai.dev
What tools or categories would you want to see reviewed? I'm especially interested in blind spots -- things I should be tracking but aren't in the catalog yet.
Top comments (0)