--- title: The Four Types of Agent Loops | daily.dev description: A newsletter covering two main topics: first, a taxonomy of four agent loop types (turn-based, goal-based, time-based, and proactive) that differ in how they... canonical: https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: The Four Types of Agent Loops | daily.dev og:description: A newsletter covering two main topics: first, a taxonomy of four agent loop types (turn-based, goal-based, time-based, and proactive) that differ in how they... og:url: https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps og:image: https://api.daily.dev/og/posts/AS3pWiEPs.png og:image:alt: The Four Types of Agent Loops og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # The Four Types of Agent Loops **[Daily Dose of Data Science \| Avi Chawla \| Substack](https://daily.dev/sources/dailydoseofds)** · 7 min read · 0 upvotes · 0 comments ## Summary A newsletter covering two main topics: first, a taxonomy of four agent loop types (turn-based, goal-based, time-based, and proactive) that differ in how they trigger runs and determine completion, helping developers choose the right structure for exploratory, measurable, recurring, or standing tasks. Second, a breakdown of NVIDIA and MIT's SparDA architecture, which adds a fourth 'Forecast' projection to transformer layers to prefetch KV cache blocks from CPU RAM ahead of time, achieving up to 1.7x faster decoding and 6.5-point accuracy gains on long-reasoning benchmarks with only 0.41% parameter overhead. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What are the four types of agent loops in loop engineering? They are turn-based, goal-based, time-based, and proactive loops. Turn-based loops are triggered by a user prompt and reviewed by a human each time. Goal-based loops run until an evaluator model confirms success criteria from a /goal command are met. Time-based loops fire on a schedule via /loop or /schedule. Proactive loops trigger on events with no human present, spawning triage, fix, and reviewer agents automatically. _daily.dev surfaces practical breakdowns like this for developers deciding how to architect agent workflows._ ### What is SparDA and how does it speed up transformer decoding? SparDA is a transformer variant from NVIDIA and MIT that adds a fourth 'Forecast' projection alongside Q, K, and V in each attention layer, predicting which KV cache blocks the next layer will need so they can be prefetched from CPU memory while the current layer computes. This decoupling delivers up to 1.7x faster decode, 1.25x faster prefill, and up to 5.3x higher decode throughput with larger batches, at a cost of only 33.5M extra parameters on an 8B model. _Developers optimizing long-context LLM inference can track architecture advances like this on daily.dev._ ### How does the mcp-use SDK let you fine-tune LLMs from Claude? The mcp-use SDK is an open-source framework for building MCP Apps that connects Claude to Hugging Face's Hub for model and dataset search, handles chat template formatting, and lets a user configure LoRA rank, quantization, batch size, and learning rate directly from a Claude conversation, with training executed on Hugging Face's GPU infrastructure via AutoTrain. Once training finishes, the fine-tuned model can be chatted with directly from Claude. _daily.dev helps developers exploring MCP-based tooling keep up with new agent app frameworks like mcp-use._ ## Similar posts on daily.dev - [Practical Loop Engineering](https://daily.dev/posts/practical-loop-engineering-5ezmrnsnj) · Elevate · 107 upvotes · 4 comments - [Your agent loop is not a production system](https://daily.dev/posts/your-agent-loop-is-not-a-production-system-s0faarjcu) · LeadDev · 0 upvotes · 0 comments - [EP215: The Anatomy of an AI Agent](https://daily.dev/posts/ep215-the-anatomy-of-an-ai-agent-0bupftjkn) · ByteByteGo · 4 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#deep-learning](https://daily.dev/tags/deep-learning), [#mcp](https://daily.dev/tags/mcp) [View this post on daily.dev](https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"The Four Types of Agent Loops","url":"https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps"},"datePublished":"2026-07-14T16:28:48.465Z","dateModified":"2026-09-14T07:02:05.631Z","description":"A newsletter covering two main topics: first, a taxonomy of four agent loop types (turn-based, goal-based, time-based, and proactive) that differ in how they...","image":"https://substackcdn.com/image/fetch/$s_!pSId!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6296b4b4-3f8c-4d3e-b193-24b456d14b03_960x922.gif","thumbnailUrl":"https://substackcdn.com/image/fetch/$s_!pSId!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6296b4b4-3f8c-4d3e-b193-24b456d14b03_960x922.gif","isAccessibleForFree":true,"articleSection":"Daily Dose of Data Science | Avi Chawla | Substack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Daily Dose of Data Science | Avi Chawla | Substack","logo":"https://media.daily.dev/image/upload/s--4IHQgTOw--/f_auto/v1710503712/logos/dailydoseofds","url":"https://daily.dev/sources/dailydoseofds"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,deep-learning,mcp","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Daily Dose of Data Science | Avi Chawla | Substack","item":"https://daily.dev/sources/dailydoseofds"},{"@type":"ListItem","position":3,"name":"The Four Types of Agent Loops"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/the-four-types-of-agent-loops-as3pwieps#faq","mainEntity":[{"@type":"Question","name":"What are the four types of agent loops in loop engineering?","acceptedAnswer":{"@type":"Answer","text":"They are turn-based, goal-based, time-based, and proactive loops. Turn-based loops are triggered by a user prompt and reviewed by a human each time. Goal-based loops run until an evaluator model confirms success criteria from a /goal command are met. Time-based loops fire on a schedule via /loop or /schedule. Proactive loops trigger on events with no human present, spawning triage, fix, and reviewer agents automatically. daily.dev surfaces practical breakdowns like this for developers deciding how to architect agent workflows."}},{"@type":"Question","name":"What is SparDA and how does it speed up transformer decoding?","acceptedAnswer":{"@type":"Answer","text":"SparDA is a transformer variant from NVIDIA and MIT that adds a fourth 'Forecast' projection alongside Q, K, and V in each attention layer, predicting which KV cache blocks the next layer will need so they can be prefetched from CPU memory while the current layer computes. This decoupling delivers up to 1.7x faster decode, 1.25x faster prefill, and up to 5.3x higher decode throughput with larger batches, at a cost of only 33.5M extra parameters on an 8B model. Developers optimizing long-context LLM inference can track architecture advances like this on daily.dev."}},{"@type":"Question","name":"How does the mcp-use SDK let you fine-tune LLMs from Claude?","acceptedAnswer":{"@type":"Answer","text":"The mcp-use SDK is an open-source framework for building MCP Apps that connects Claude to Hugging Face's Hub for model and dataset search, handles chat template formatting, and lets a user configure LoRA rank, quantization, batch size, and learning rate directly from a Claude conversation, with training executed on Hugging Face's GPU infrastructure via AutoTrain. Once training finishes, the fine-tuned model can be chatted with directly from Claude. daily.dev helps developers exploring MCP-based tooling keep up with new agent app frameworks like mcp-use."}}]} ```