Archive
Discover and discuss technology tools
Explore the Tiscuss archive by category or keyword, then jump into conversations around what matters most.
How AI Will Transform Office Communication
How will work setups change if we spend more and more time talking to our computers?
AI Safety Measures: Controlling AI Agents' Destructive Actions
Saw a case recently where an AI coding agent ended up wiping a database in seconds. It made me think about how most agent setups are wired: agent decides → executes query → done There’s usually logging-tracing but those all happen after the action. If your agent has access to systems like a DB, are you: restricting it to read-only? running everything in staging/sandbox? relying on prompt-level safeguards? or putting some kind of control layer in between?
Trading System V2: AI's Role in Deterministic Execution
Thanks to the incredible feedback on my last post, I’m officially moving away from the "distributed veto" system (where 8 LLM agents argue until they agree to trade). For v2, I am implementing a strict State Machine using a deterministic runtime (llm-nano-vm). The new rule is simple: Python owns the math and the execution contract. The LLM only interprets the context. I've sketched out a 5-module architecture, but before I start coding the new Python feature extractors, I want to sanity-check the exact roles I’m giving to the AI. Here is the blueprint: 1. The HTF Agent (Higher Timeframe - D1/H4) Python: Extracts structural levels, BOS/CHoCH, and premium/discount zones. LLM Role: Reads this hard data to determine the institutional narrative and select the most relevant Draw on Liquidity (DOL). 2. The Structure Agent (H1) Python: Identifies all valid Order Blocks (OB) and Fair Value Gaps (FVG) with displacement. LLM Role: Selects the highest-probability Point of Interest (POI) based on the HTF Agent's narrative. 3. The Trigger Agent (M15/M5) 100% Python (NO LLM): Purely deterministic. It checks for liquidity sweeps and LTF CHoCH inside the selected POI. 4. The Context Agent LLM Role: Cross-references active killzones, news blackouts, and currency correlations to either greenlight or veto the setup. 5. The Risk Agent 100% Python (NO LLM): Calculates Entry, SL, TP, Expected Value (EV), and position sizing. The state machine will only transition to EXECUTING if the deterministic Trigger and Risk modules say yes. The LLMs are basically just "context providers" for the state machine. My questions for the quants/architects here: Does this division of labor make sense? Am I giving the LLMs too much or too little responsibility in step 1 and 2? By making the Trigger layer (M15/M5) 100% deterministic, am I losing the core advantage of having an AI, or is this the standard way to avoid execution paralysis? Would you merge the HTF and Structure agents to reduce token constraints/hallucinations, or is separating them better for debugging? Would love to hear your thoughts before I dive into the codebase.
Explore Agentic AI with Free Interactive Curriculum on AgentSwarms
Hey Everyone, Over the last few months, I noticed a massive gap in how we learn about Agentic AI. There are a million theoretical blog posts and dense whitepapers on RAG, tool calling, and swarms, but almost nowhere to just sit down, run an agent, break it, and see how the prompt and tools interact under the hood. So, I built **AgentSwarms**.fyi It’s a free, interactive curriculum for Agentic AI. Instead of just reading, you run live agents alongside the lessons. **What it covers:** * Prompt engineering & system messages (seeing how temperature and persona change behavior). * RAG (Retrieval-Augmented Generation) vs. Fine-tuning. * Tool / Function Calling (OpenAI schemas, MCP servers). * Guardrails & HITL (Human-in-the-Loop) for safe deployments. * Multi-Agent Swarms (orchestrators vs. peer-to-peer handoffs). **The Tech/Setup:** You don't need to install anything or provide API keys to start. The "Learn Mode" is completely free and sandboxed. If you want to mess around with your own models, there's a "Build Mode" where you can plug in your own keys (OpenAI, Anthropic, Gemini, local models, etc.). I’d love for this community to tear it apart. What agent patterns am I missing? Is the observability dashboard actually useful for debugging your traces? Let me know what you think.
Arc Gate: Advanced Prompt Injection Protection for OpenAI
Built Arc Gate — sits in front of any OpenAI-compatible endpoint and blocks prompt injection before it reaches your model. Try it here — no signup, no code, no setup: https://web-production-6e47f.up.railway.app/try Type any prompt and see if it gets blocked or passes. The examples on the page show the difference. The main detection layer is a behavioral SVM on sentence-transformer embeddings — catches semantic intent, not just pattern matches. Phrase matching is just the fast first pass. Four layers total. Benchmarked on 40 OOD prompts (indirect, roleplay, hypothetical framings — the hard stuff): • Arc Gate: Recall 0.90, F1 0.947 • OpenAI Moderation: Recall 0.75, F1 0.86 • LlamaGuard 3 8B: Recall 0.55, F1 0.71 Zero false positives on benign prompts including security discussions and safe roleplay. Block latency 329ms. One URL change to integrate into your own project: base\_url=“https://web-production-6e47f.up.railway.app/v1” GitHub: github.com/9hannahnine-jpg/arc-gate — star if useful.
AI Tool Locus: Autonomous Business Operations
This sub has seen enough "AI can now do X" posts to have a finely tuned radar for what's real and what's a demo that falls apart the moment someone actually uses it. So I'll skip the hype and just tell you what we built and where the edges are. The core problem we were solving wasn't any individual capability. Generating copy is solved. Building websites is solved. Running ads is mostly solved. The unsolved problem was coherent autonomous decision making across all of those systems simultaneously without a human acting as the integration layer between them. That's what we spent most of our time on. Locus Founder takes someone from idea to fully operational business without them touching a single tool. The system scopes the business, builds the infrastructure, sources products, writes conversion optimized copy, and then runs paid acquisition across Google, Facebook and Instagram autonomously. Continuously. Not as a one time setup but as an ongoing operation that monitors performance and adjusts without being told to. The honest version of where AI actually performs well in this system and where it doesn't: It's genuinely good at the build layer. Storefront generation, copy, pricing structure, initial ad creative, coherent and fast in a way that would have been impossible two years ago. The operations layer is more complicated. Autonomous ad optimization works well within normal parameters. The judgment calls that fall outside those parameters, unusual market conditions, supplier issues, platform policy edge cases, are still the places where the system makes decisions a human would immediately recognize as wrong. That gap between capability and judgment is the most interesting unsolved problem in what we're building and probably in the agent space generally right now. We got into YCombinator this year. Opening 100 free beta spots this week before public launch. Free to use, you keep everything you make. For people in this sub specifically, less interested in the "wow AI can do that" reaction and more interested in people who want to actually stress test where the judgment breaks down. Beta form: [https://forms.gle/nW7CGN1PNBHgqrBb8](https://forms.gle/nW7CGN1PNBHgqrBb8) Where do you think autonomous business judgment actually gets solved and what does that look like?
Explore AgentSwarms: Free Hands-On Learning for Agentic AI
Explore AgentSwarms: Free Hands On Learning for Agentic AI Discover the Power of Agentic AI with AgentSwarms AgentSwarms is an innovative platform designed to o…
AI Agents: Identity, Not Memory, Was the Key to Stability
Everyone's building memory layers right now. Longer context, better embeddings, persistent state across sessions. I spent weeks on the same thing. But the failure mode that actually cost me the most debugging time had nothing to do with memory. Here's what it looked like: an agent would be technically correct - good reasoning, clean output - but operating from the wrong context entirely. Answering questions nobody asked. Taking actions outside its scope. Not hallucinating. Drifting. Like a competent person who walked into the wrong meeting and started contributing without realizing they're in the wrong room. I run 11 persistent agents locally. Each one is a domain specialist - its entire life is one thing. The mail agent's every session, every test, every bug fix is about routing messages. The standards auditor's whole existence is quality checks. They're not generic workers configured for a task. They've each accumulated dozens of sessions of operational history in their domain, and that history is what makes them good at their job. When they started drifting, my first instinct was what everyone's instinct is: better memory. More context. None of it helped. An agent with perfect recall of its last 50 sessions would still lose track of who it was in session 51. What actually fixed it I separated identity from memory entirely. Three files per agent: passport.json - who you are. Role, purpose, principles. Rarely changes. This is the anchor. local.json - what happened. Rolling session history, key learnings. Capped and trimmed when it fills up. observations.json - what you've noticed about the humans and agents you work with. Concrete stuff like "the git agent needs 2 retries on large diffs" or "quality audits overcorrect on technical claims." The agent writes these itself based on what actually happens. Identity loads first, then memory, then observations. That ordering matters. When the identity file loads first, the agent has a stable reference point before any history lands. The mail routing agent learned the sharpest version of this. When identity was ambiguous, it would route messages from the wrong sender. The fix wasn't better routing logic - it was: fail loud when identity is unclear. Wrong identity is worse than silence. The files alone weren't enough Three JSON files helped, but didn't scale past a few agents. What actually made 11 work is that none of them need to understand the full system. Hooks inject context automatically every session - project rules, branch instructions, current plan. One command reaches any agent. Memory auto-archives when it fills up. Plans keep work focused so agents don't carry their entire history in context. The system learned from failing. The agents communicate through a local email system - they send each other tasks, status updates, bug reports. One agent monitors all logs for errors. When it spots something, it emails the agent who owns that domain and wakes them up to investigate. The agents fix each other. The memory agent iterated three sessions to fix a single rollover boundary condition - each time it shipped, observed a new edge case, and improved. These aren't cold modules. They break, they help each other fix it, they get better. That's how the system got to where it is. You don't need 11 agents The 11 agents in my setup maintain the framework itself. That's the reference implementation. But u could start with one agent on a side project - just identity and memory, pick up where u left off tomorrow. Need a team? Add a backend agent, a frontend agent, a design researcher. Three agents, same pattern, same commands. Or scale to 30 for a bigger system. Each new agent is one command and the same structure. What this doesn't solve This all runs locally on one machine. I don't know whether identity drift looks the same in hosted environments. If u run stateless agents behind an API, the problem might not exist for you. Small project, small community, growing. The pattern itself is small enough to steal - three JSON files and a convention. But the system that keeps agents coherent at scale is where the real work went. pip install aipass and two commands to get a working agent. The .trinity/ directory is the identity layer. Has anyone else tried separating identity from memory in their agent setups? Curious whether the ordering matters in other architectures, or if it's just an artifact of how this system evolved.
Explore AgentSwarms: Free Hands-On Agentic AI Learning
Explore AgentSwarms: Free Hands On Agentic AI Learning Introduction AgentSwarms offers a unique, hands on approach to learning about Agentic AI, providing a fre…