Co-Founder and CPO at Qodo
Joined January 2015
- Tweets114
- Following67
- Followers268
- Likes129
Pinned Tweet
๐ Introducing AlphaCodium by @CodiumAI :
A new state-of-the-art AI code generation system for competitive programming, fully open-sourced! ๐ง ๐ป
๐ Transforms code generation quality through an iterative, test-based refinement process.
๐ Boosts GPT-4's code accuracy from 19% to 44% on CodeContests (validation set).
๐ Explore, contribute, and innovate:
Blog: codium.ai/blog/alphacodium-sโฆ
Paper: arxiv.org/abs/2401.08500
Open-source project: github.com/Codium-ai/AlphaCoโฆ
Dedy Kredo retweeted
Today I was fired from OpenAI.
I was in charge of login and voice integration in Codex.
Used codex /review and /verify.
Both passed. I slept beautifully.
In hindsight, only one of us could get fired if something goes wrong, and it wasnโt Codex.
My next career move: becoming a VC backing startups that build code governance, quality, and verification tools.
Iโll be reviewing your decks with Codex.
Dedy Kredo retweeted
Weโre all building our own software factories.
But, our coding and code review agents are missing a curated map of the software theyโre working on: how repositories connect, what depends on what, and where a change can have an impact.
Thatโs why we built Software Map in Qodo ๐
Install Qodo and get your software map within hours.
Watch the video for examples and a closer look.
AI can generate code faster than ever. That part is solved. The real challenge now is trust.
@QodoAI just raised $70M in Series B ($120M total). The same AI that writes your code can't be the one governing its quality. That's a structurally different problem.
We built that different system. Ranked #1 in independent code review benchmarks. Trusted by NVIDIA, Walmart, Red Hat, Intuit, and more. 11x enterprise growth this year.
Intelligence is enough for generation. Wisdom is a must for governance
Full demo + blog below.
qodo.ai/blog/qodo-70m-seriesโฆ
Dedy Kredo retweeted
Qodo ranks FIRST with F1 score on @withmartian's independent AI Code Review Benchmark.
The result reflects what we're building: an AI code quality and governance platform engineered for the enterprise.
Exhaustive (default) results in the @QodoAI Code Review Benchmark โ same setup/dataset/pipeline:
Sonnet 4.6: 82% precision, 59% recall, 69 F1
GPT 5.2: 74% precision, 61% recall, 67 F1
Opus 4.6: 85% precision, 55% recall, 67 F1
Sonnet 4.5: 78% precision, 56% recall, 65 F1
๐ Biggest jump: Sonnet 4.6 vs 4.5 = +4 precision, +3 recall, +4 F1.
Note: this is the main review model - Qodo also combines other models across the full system.
Most teamsโ โcoding standardsโ are scattered across old docs, repo files, and various coding agentsโ Rules/Skills files โ plus tribal knowledge.
Qodo 2.1 ships a Rules System that fixes this:
- learns rules from your codebase + PR history
- suggests + lets you edit/approve
- enforces them in every PR (deterministic, not vibes)
- dedupes/conflict-checks as things evolve
- measures impact (adoption + violations)
One source of truth. Enforced everywhere. Evolving with your code. ๐
@AnthropicAI Opus 4.6 > Opus 4.5 on the @QodoAI Code Review Benchmark - across both review modes:
Exhaustive (higher recall)
โข F1: 60.2 vs 57.1 (+3.1)
โข Recall: 51.1 vs 46.5 (+4.6)
Precise (higher precision)
โข F1: 58.9 vs 52.4 (+6.5)
โข Recall: 46.0 vs 38.8 (+7.2)
Better tradeoff, less missed issues.
Dedy Kredo retweeted
Qodo Context Engine agents (Deep Research & PR Knowledge) are now smarter after experimenting with @OpenAI 's GPT-5.1-Codex-Max.
Beating our internal benchmarks by a wide margin, Qodo context engine now has deeper understanding, accurate answers, and is a smarter teammate.
โก๏ธ As fast and affordable as GPT-5.1
๐ง Much smarter with minimal token overhead
โ๏ธ Handles custom tools better than any LLM we've tested
Here's our benchmark showing recall vs number of tokens & steps (Based on DeepCodeBench evaluation)
qodo.ai/blog/deepcodebench-rโฆ
Dedy Kredo retweeted
This is insane power for software developers:
An agentic principal engineer that you can install as an MCP server. It can index any codebase and surface insights you didn't even know about.
This is likely the most useful MCP server you'll start using.
It's called @Qodo Aware, and it's the first production-ready deep research agent built for navigating and understanding enterprise-scale codebases.
Find the GitHub repository in a comment below.
The beauty of this is that you don't have to switch tools to take advantage of it: you can keep using Cursor, Visual Studio Code, Claude Code, or whatever else you are using.
This is an MCP server, so you can simply connect your agent to it!
This is a goated tool, and here is why:
โข It can trace how a feature works end-to-end
โข It can map dependencies across the entire codebase
โข It can trace bugs across multiple repositories
โข It can check the impact of changes across services
โข It can plan large architectural changes across repositories
โข It can answer questions about your code and dependencies
โข It can surface relevant context during code reviews
โข It can spot cross-system issues in the repository
Dedy Kredo retweeted
๐Announcing Qodo Aware ๐
Your AI coding assistant is flying blind. It sees a few hops from your current file, but misses a LOT of context that matters.
Qodo Aware is the first and ONLY deep research agent built for finding the right context across large, complex codebases: 10 repos or 1,000. ๐ฅ
Read the blog here -> qodo.ai/blog/introducing-qodโฆ
Dedy Kredo retweeted
Catch @DedyKredo live on todayโs stream!
Dedy Kredo retweeted
GPT-5 is here and itโs OpenAI day.
On todayโs stream:
โ@markchen90 (OpenAI)
โ@gdb (OpenAI)
โ@thefriley (OpenAI)
โ@bradlightcap (OpenAI)
โ@max_a_schwarzer (OpenAI)
โ@ScottWu46 (Cognition)
โ@DedyKredo (Qodo)
โ@zachlloydtweets (Warp)
โ@rileytomasek (Charlie Labs)
โ@rauchg (Vercel)
โ@EnoReyes (Factory)
โ@guygr (Augment)
โ@harjotsgill (CodeRabbit)
โ@clairevo (ChatPRD)
โ@benhylak (Raindrop)
See you all on the show.
Dedy Kredo retweeted
2025 is the year of the CLI,
not ๐ฅ๐ฆ๐ด๐ฑ๐ช๐ต๐ฆ its simplicity but ๐ฃ๐ฆ๐ค๐ข๐ถ๐ด๐ฆ of it.
Iโm thrilled to share todayโs announcement.
Itโs a milestone for Qodoโs mission, and for every team that wants to put AI to work across the SDLC.
Letโs dive in!
๐๐๐๐ญ ๐๐จ๐๐จ ๐๐๐ง ๐๐๐
Built for creating, running, and managing AI agents that automate critical development tasks across the SDLC, from analyzing CI tests and logs to triaging production errors and more.
๐๐๐ฒ ๐๐๐๐ญ๐ฎ๐ซ๐๐ฌ
โ
Runs everywhere
โ
Designed for task execution, not just Q&A
โ
Trigger from CI, hooks, or cloud functionsโyou choose
โ
Works with all major models
โ
Fully customizable agents with their own tools, credentials, and goals
And yes, it plays nicely with our other Qodo agents too (e.g., Qodo Merge, the AI code-review tool).
This isnโt just about writing codeโitโs about catching bugs earlier, automating deeper, and shipping higher-quality software.
๐๐ซ๐ฒ ๐ข๐ญ ๐ญ๐จ๐๐๐ฒ ๐ข๐ง ๐ฌ๐๐๐จ๐ง๐๐ฌ
Make sure you have npm installed, then run:
npm install -g @qodo/gen
Oh, one more thing-
๐๐ฑ๐ฉ๐ฅ๐จ๐ซ๐ ๐จ๐ฎ๐ซ ๐ฅ๐ข๐๐ซ๐๐ซ๐ฒ ๐จ๐ ๐ฌ๐ฉ๐๐๐ข๐๐ฅ๐ข๐ณ๐๐ ๐๐จ๐๐จ ๐๐ ๐๐ง๐ญ๐ฌ,
and contribute your own so others can benefit!
github.com/qodo-ai/agents
Dedy Kredo retweeted
๐จ Just dropped: Qodo Gen CLI
Bring AI agents to your terminal โ and your entire SDLC.
No IDE required. No context lost.
โ
Build custom agents for your stack
๐ Run workflows from commit to deploy
๐ Background agents that track your code
๐ Git-aware. CI-ready. Terminal-native
๐ฅ Watch the demo ๐
Our team at @QodoAI tested @OpenAI's new GPT-4.1 on a real-world coding task: providing code review for actual pull requests.
After running it through 200 PRs, hereโs what stood out:
GPT 4.1 was better than Claude 3.7 Sonnet 55% of the time.
- It provided more focused, precise, and actionable feedback
- It caught real bugs like off-by-one errors and invalid object spreads
- It knew when to be thorough, and when to stay quiet
Our results are discussed in more detail in the official OpenAI blog post about GPT 4.1 and in our blog post.
P.S. GPT-4.1 is supported in both of our products - Qodo Merge and Qodo Gen!
qodo.ai/blog/benchmarked-gptโฆ
openai.com/index/gpt-4-1/
