Show HN: Statewright – Visual state machines that make AI agents reliable https://ift.tt/mF56CbG

Show HN: Statewright – Visual state machines that make AI agents reliable Agentic problem solving in its current state is very brittle. I fell in love with it, but it creates as many problems as it solves. I'm Ben Cochran, I spent 20+ years in the trenches with full-stack Engineering, DevOps, high performance computing & ML with stints at NVIDIA, AMD and various other organizations most recently as a Distinguished Engineer. For agents to work reliably you either need massive parameter counts or massive context windows to keep the solution spaces workable. Most people are brute forcing reliability with bigger models and longer prompts. What if I made the problem smaller instead of making the model bigger? I took a different approach by using smaller models: models in the 13-20B parameter range and set them to task solving real SWE-bench problems. I constrained the tool and solution spaces using formal state machines. Each state in the machine defines which tools the model can access, how many iterations it gets and what transitions are valid. A planning state gets read-only tools. An implementation state gets edit tools (scoped to prevent mega edits) and write friendly bash tools. The testing state gets bash but only for testing commands. The model cannot physically skip steps or use the wrong tool at the wrong time. It is enforced via protocol, not via prompts. The results were more promising than I would have expected. Across multiple model families irrespective of age (qwen-coder, gpt-oss, gemma4) and the improvements were consistent above the 13B parameter inflection point. Below that, models can navigate the state machine but can't retain enough context to produce accurate edits. More on the research bit: https://ift.tt/4omOiAX Surprisingly this yielded improvements in frontier models as well. Haiku and Sonnet start to punch above their weight and Opus solves more reliably with fewer tokens and death spirals. Fine tuning did not yield these kinds of functional improvements for me. The takeaway it seems is that context window utilization matters more than raw context size - a tightly scoped working context at each step outperforms a model given carte blanche over everything. Constraining LLMs which are non-idempotent by using deterministic code is a pattern that nobody is currently talking about. So, I built Statewright. Its core is a Rust engine that evaluates state machine definitions: states, transitions, guards and tool restrictions. Its orchestration doesn't use an LLM, just enforces the state machine. On top of that is a plugin layer that integrates with Claude Code (and soon Codex, Cursor and others) via MCP. When you activate a workflow, hooks enforce the guardrails per state automatically. The model sees 5 tools available instead of dozens, gets clear instructions for the current phase and transitions when conditions are met. Importantly it tells the model when it's attempting to do something that isn't in scope, incorrect or when it needs to try something else after getting stuck. You can use your agent via MCP to build a state machine for you to solve a problem in your current context. The visual editor at statewright.ai lets you tweak these workflows in a graph view... You can clearly see the failure paths, the retry loops and the approval gates. State machines aren't DAGs; they loop and retry, which is what agentic work actually needs. Statewright is currently live with a free tier, try it out in Claude Code by running the following: /plugin marketplace add statewright/statewright /plugin install statewright /reload-plugins Then "start the bugfix workflow" or /statewright start bugfix. You'll need to paste your API key when prompted. The latest versions of Claude may complain -- paste the API key again and say you really mean it, Claude is just being cautious here. Feedback is welcome on the workflow editor, the plugin experience, and tell me what workflows you'd want to build first. Agents are suggestions, states are laws. https://ift.tt/aAXUYbq May 12, 2026 at 07:54PM

Show HN: Mimik – open-source local-first alternative to Scribe and Tango https://ift.tt/sH6jgRt

Show HN: Mimik – open-source local-first alternative to Scribe and Tango https://ift.tt/0t4mjkF May 11, 2026 at 11:18PM

Show HN: SyncBank – Self-hosted bank sync for EU banks https://ift.tt/XemyZDk

Show HN: SyncBank – Self-hosted bank sync for EU banks https://syncbank.app/ May 11, 2026 at 11:32PM

Show HN: adamsreview – better multi-agent PR reviews for Claude Code https://ift.tt/XBinFLg

Show HN: adamsreview – better multi-agent PR reviews for Claude Code I built adamsreview, a Claude Code plugin that runs deeper, multi-stage PR reviews using parallel sub-agents, validation passes, persistent JSON state, and optional ensemble review via Codex CLI and PR bot comments. On my own PRs, it has been catching dramatically more real bugs than Claude’s built-in /review, /ultrareview, CodeRabbit, Greptile, and Codex’s built-in review, while producing fewer false positives. adamsreview is six Claude Code slash commands packaged as a plugin: review, codex-review, add, promote, walkthrough, and fix. I modeled it after the built-in /review command and extended it meaningfully. You can clear context between review stages because state is stored in JSON artifacts on disk, with built-in scripts for keeping it updated. The walkthrough command uses Claude’s AskUserQuestion feature to walk you through uncertain findings or items needing human review one by one. Then, the fix command dispatches per-fix-group agents and re-reviews the work with Opus, reverting any regressions before committing survivors. It runs against your regular Claude Code subscription (Max plan recommended), unlike /ultrareview, which charges against your Extra Usage pool. I would love feedback from Claude Code users, pro devs, and anyone with strong opinions about AI code reviews. Repo: https://ift.tt/VT81R0X Install: /plugin marketplace add adamjgmiller/adamsreview, /plugin install adamsreview@adamsreview https://ift.tt/VT81R0X May 11, 2026 at 07:36AM

Show HN: I trained a chess engine to play like humans https://ift.tt/nHYXWtr

Show HN: I trained a chess engine to play like humans I built 1e4.ai - a chess web app where you play against neural networks trained to mimic human Lichess players at specific Elo ranges. There's a separate model for each 100-point rating bucket from ~800 to 2200+, and the bots not only choose human-like moves but also burn clock time, play worse under time pressure, and blunder in human-like ways. Live demo: https://1e4.ai Code: https://ift.tt/NmLVlrv A few things that might be interesting: - Trained on almost a full year of Lichess blitz games, around 1B total games - Architecture is an a small (~9MM parameters) transformer-based network that takes the board, recent move history, the player's rating, and remaining clock time as input. Three separate models per rating bucket: move, clock-usage, and win probability. The clock model is what makes the bots feel humanish under time pressure rather than instant. Because the move model takes the clock as one input parameter, it also learns to blunder under time pressure like a human might. - Because the network is so tiny, no GPU is needed for inference - it runs easily on a local CPU - Downside of the tiny network is that it's a bit weak as you turn up the rating past around 1700. It can spot short tactics but not long multi-move combinations. - Initial training on a rented 8xH100 cluster, then fine-tunes on my local GPU for different rating ranges - Inspired by Maia-2 and DeepMind's "Grandmaster-Level Chess Without Search". On a held-out Lichess blitz benchmark, the it beats Maia-2 blitz on top-1 move prediction (56.7% vs 52.7%) and pretty substantially on win-probability calibration (Brier 0.176 vs 0.272). Numbers and code in https://ift.tt/SFe81RY... - The data pipeline is C++ via nanobind, then training with Pytorch. Getting this right was actually the thing I spent the most time on. Pre-shuffling the dataset and then being able to read the shuffled dataset sequentially at training time kept the GPU utilization high. Without this it spent a huge percentage of time on I/O while the GPU sat idle. Happy to answer questions about the rating-conditioning, the clock model, or the data pipeline. May 11, 2026 at 04:01AM

Show HN: Hustler Bingo – a tiny bingo game about startup Twitter clichés https://ift.tt/yRxaJoi

Show HN: Hustler Bingo – a tiny bingo game about startup Twitter clichés I built this after my brother started complaining that I got too much into brainrot culture. It's just for fun nothing serious, but was able to test vercel, tanstack start and convex without high stakes. Have fun! This is the game where lower score is goood for your mental health https://ift.tt/RDY4bMJ May 11, 2026 at 02:06AM

Show HN: Mosaic – arrange iOS icons by color using an evolutionary algorithm https://ift.tt/xy6NF3v

Show HN: Mosaic – arrange iOS icons by color using an evolutionary algorithm It started out as a way for me to freshen up my C++ skills during COVID. But life got in the way and it was put on ice. Luckily, coding LLMs came to the rescue and allowed me to bring it to a point where I feel comfortable sharing it. https://ift.tt/bcDNlUp May 10, 2026 at 11:59PM