IdeaWell

Project ideas from Hacker News discussions.

Agents need control flow, not more prompts

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

4 DominantThemes

Theme Core idea Supporting quotation
1. LLMs are fundamentally non‑deterministic You cannot expect a language model to guarantee that a particular result will occur; reliability can’t be built on its deterministic behavior. “If you're trying to get reliability and determinism out of the LLM, you've already lost” (Neywiny)
2. Deterministic scaffolding is required for real reliability Scripts, hooks, or small deterministic programs are the main way to anchor LLM output and make it cost‑effective. “Determinism is a different matter. Scripts and hooks are really the main levers you can pull there, but yeah - a decent script and a cron job will handle certain things much better (and for a fraction of the cost)” (bwestergard)
3. Shift from prompt‑only to programmatic control flow The real productivity gain comes from moving the heavy lifting of execution out of the prompt into a deterministic harness; LLMs become a decision‑making layer only. “The breakthrough in ai coding was not that AI intelligence increased as much as that a lot of the core process execution moved out of the LLM prompt and into the harness.” (eth415)
4. Skepticism toward full‑blown agent frameworks Most tasks don’t need a complex agent; many argue the current wave of “agents” is over‑engineered, token‑expensive, and needs stronger verification. “Agents aren't reliable; use workflows instead.” (tim‑projects)

Summary – The discussion clusters around the impossibility of pure LLM determinism, the necessity of surrounding deterministic code/harnesses, a migration from prompting to programmatic control, and a growing weariness of elaborate agent architectures that lack reliable safeguards.

🚀 Project Ideas

AgentGuard

Summary

  • Provides a declarative, auditable harness that forces LLMs to follow a strict execution flow and guarantees deterministic output validation.
  • Turns non‑deterministic prompt‑driven agents into reproducible pipelines with built‑in guards, reducing costly hallucinations and re‑runs.

Details| Key | Value |

|-----|-------| | Target Audience | Engineering teams building AI‑augmented CI/CD, data‑processing pipelines, or internal automation tools. | | Core Feature | Deterministic workflow engine that enforces pre‑defined steps, validates each step with custom tests, and logs an immutable execution receipt. | | Tech Stack | Rust backend, WebAssembly front‑end, PostgreSQL for audit logs, OpenAPI spec for integrations. | | Difficulty | Medium | | Monetization | Revenue-ready: Enterprise SaaS tiered pricing (starting at $49 / month per seat). |

Notes

  • HN commenters repeatedly cite the need for “deterministic control flow” and “verification layers” – AgentGuard directly addresses those.
  • Potential for wide discussion around replacing fragile prompt‑only approaches with enforceable pipelines; practical utility in production‑grade automation.

SkillPrompt Compiler (SPC)

Summary

  • Automatically transforms high‑level orchestration prompts (e.g., “run a validator, then apply the patch, then test”) into version‑controlled scripts that can be reused and audited.
  • Eliminates the “prompt‑as‑code” anti‑pattern by compiling declarative intent into deterministic, testable artifacts.

Details

Key Value
Target Audience Solo developers and small dev‑ops teams who currently maintain ad‑hoc prompt scripts in Markdown or JSON.
Core Feature Compiler that ingests a YAML/JSON orchestration brief and emits a self‑contained Python or Bash script with embedded call‑to‑LLM wrappers and validation hooks.
Tech Stack Node.js (compiler), Python runtime, Git hooks for CI validation, Docker for isolated execution.
Difficulty Low
Monetization Hobby

Notes

  • Directly resonates with HN users who emphasize “write code to do deterministic work” – the tool automates that workflow creation.
  • Sparks conversation about reducing token waste and improving reliability; practical for personal projects and early‑stage startups.

GuardianAngel

Summary

  • Multi‑agent verification layer that cross‑checks LLM‑generated outputs against a set of unit‑test‑style contracts, rejecting non‑conforming results before they reach production.
  • Provides a lightweight “trust but verify” safety net for any AI‑driven code‑generation pipeline.

Details

Key Value
Target Audience QA engineers, security‑focused dev teams, and SaaS platforms that expose AI‑generated outputs to customers.
Core Feature Contract‑driven validation engine: users define JSON‑Schema or Rust traits; GuardianAngel runs generated artifacts through a suite of deterministic tests and returns pass/fail with a detailed report.
Tech Stack Go microservice, JSON‑Schema validator, Rust for contract definitions, Prometheus metrics for exposure monitoring.
Difficulty High
Monetization Revenue-ready: Revenue-ready: SaaS with usage‑based pricing (e.g., $0.01 per validation call).

Notes

  • Directly answers HN concerns about “hallucination” and “unreliable output” – an automated guardrail that catches errors before deployment.
  • Likely to generate rich discussion on reliability engineering for AI systems and offers immediate practical utility for any AI‑centric CI pipeline.

DAGFlow

Summary

  • Open‑source workflow engine that natively blends deterministic DAG execution with on‑demand LLM tool calls, enabling complex multi‑step AI tasks while preserving full traceability.
  • Makes it trivial to embed “if‑else” logic, parallel branches, and loop constructs around probabilistic AI components.

Details

Key Value
Target Audience Platform engineers building sophisticated AI‑enhanced automations (e.g., research assistants, data‑pipeline orchestrators).
Core Feature Graph‑based workflow definition language (YAML/JSON) that supports nodes for “LLM call”, “script execute”, “conditional”, and “join”; runtime enforces deterministic ordering and resource limits.
Tech Stack Python (networkx for DAG), FastAPI for REST endpoints, SQLite for state persistence, Grafana for observability.
Difficulty Medium
Monetization Hobby

Notes

  • Aligns with the “control flow tells the agent what it’s allowed to do” theme prevalent on HN; provides an actual control‑flow layer rather than just prompts.
  • Sparks debate on the trade‑offs between full‑stack orchestration frameworks and lightweight skill‑based approaches; practical for building production‑grade AI pipelines.

ResourceGuard API Gateway#Summary

  • A managed API gateway that throttles, caps, and enforces deterministic quotas on LLM calls, ensuring that stochastic outputs are bounded in cost, latency, and side‑effects.
  • Couples quota enforcement with automatic fallback deterministic logic when thresholds are exceeded.

Details| Key | Value |

|-----|-------| | Target Audience | Product managers and SaaS operators who expose LLM‑powered features to end users and need predictable billing and reliability. | | Core Feature | Rate‑limiting, token‑budget enforcement, and automatic fallback routing to pre‑written scripts or rule‑based services when the LLM budget is exhausted. | | Tech Stack | Node.js gateway, Redis for token accounting, OpenTelemetry for tracing, Kubernetes for scaling. | | Difficulty | Medium | | Monetization | Revenue-ready: Tiered pricing based on monthly token quota (e.g., $0.001 per 1 K tokens, with enterprise caps). |

Notes- Directly tackles the “token cost” and “unreliable outcome” concerns highlighted in many HN threads; offers a concrete service model.

  • Generates discussion on sustainable AI product design and provides an immediately deployable solution for companies looking to offer AI features without runaway expenses.

Read Later