--- title: Engineering Multi-Agent AI Teams That Build and Test... description: Salesforce's Marketing Cloud engineering team built Agent Designer, a system where an orchestrator and seven specialist AI agents design, verify, test, and... canonical: https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Engineering Multi-Agent AI Teams That Build and Test Themselves | daily.dev og:description: Salesforce's Marketing Cloud engineering team built Agent Designer, a system where an orchestrator and seven specialist AI agents design, verify, test, and... og:url: https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11 og:image: https://api.daily.dev/og/posts/EhCIxev11.png og:image:alt: Engineering Multi-Agent AI Teams That Build and Test Themselves 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. # Engineering Multi-Agent AI Teams That Build and Test Themselves **[Salesforce Engineering](https://daily.dev/sources/salesforceeng)** · 7 min read · 1 upvotes · 0 comments ## Summary Salesforce's Marketing Cloud engineering team built Agent Designer, a system where an orchestrator and seven specialist AI agents design, verify, test, and repair other multi-agent teams, cutting build time from two-to-four hours down to 15-30 minutes. The orchestrator itself is barred from writing agent files, enforcing a strict separation between planning and execution. The Q&A covers how the team resolved conflicting architecture, failure-mode, and compliance decisions; enforced per-topology budget formulas; made agent verification machine-readable with PASS/WARN/FAIL contracts plus a meta-check across verifiers; and bounded self-healing repair loops to two fix attempts and a $3 spending ceiling to prevent runaway or misdirected fixes. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How do you prevent AI agents from repeatedly trying to fix the wrong bug in a self-healing repair loop? Bound the retry process with hard limits: Salesforce's Agent Designer restricts self-healing to two fix attempts and a $3 spending ceiling per team. A test doctor first classifies the failure type (violation, permission block, tool error, boot error, soft failure, infinite loop, or state corruption) before attempting repair, and the system escalates to a human instead of continuing to guess after limits are hit. _Anyone building autonomous repair loops can compare bounded-retry strategies like this one on daily.dev._ ### How do you keep AI verifier agents from either rubber-stamping bad designs or blocking good ones with false failures? Require verifiers to output structured PASS, WARN, or FAIL results instead of free-form text, then run a meta-check comparing their outputs against each other and the proposed design. Rules like warning rather than failing when uncertain, and not failing a design solely for a specific permission flag, help distinguish real violations from reviewable concerns, since no single verifier is trusted in isolation. _Teams designing agent verification pipelines can track patterns like this on daily.dev while refining their own checks._ ### How should you calculate spending budgets for different multi-agent architectures like swarms versus single agents? Apply a different formula per topology instead of one static limit: a flat check for a single agent, items multiplied by cost multiplied by steps for a swarm, and whole-roster ceilings for a team. Salesforce's Agent Designer stores these budgets, turn limits, and cost ceilings centrally, and also divides an orchestrator's own 500-turn budget across analysis, approval, generation, testing, and repair phases. _Developers weighing budget controls across agent topologies can follow similar orchestration approaches on daily.dev._ ## Similar posts on daily.dev - [How Agentforce Enhanced Chat Built an Agent-first Chat Experience](https://daily.dev/posts/how-agentforce-enhanced-chat-built-an-agent-first-chat-experience-szihcoqon) · Salesforce Engineering · 1 upvotes · 0 comments - [The Future of the Salesforce Developer in the Age of Agentic AI](https://daily.dev/posts/the-future-of-the-salesforce-developer-in-the-age-of-agentic-ai-au2kgehl9) · Salesforce Developers Blog · 1 upvotes · 0 comments - [From a Jira Ticket to a Running Salesforce UI Test, Written by an AI](https://daily.dev/posts/from-a-jira-ticket-to-a-running-salesforce-ui-test-written-by-an-ai-un9gyvfgm) · Medium · 1 upvotes · 0 comments - [“Six tools, one harness”: Salesforce loops together a six-pack of favorites](https://daily.dev/posts/six-tools-one-harness-salesforce-loops-together-a-six-pack-of-favorites-a936j1oy5) · The New Stack · 1 upvotes · 0 comments - [Agentic AI for Test Workflows. Why Our QA Team Built It and How Testing Changed as a Result](https://daily.dev/posts/agentic-ai-for-test-workflows-why-our-qa-team-built-it-and-how-testing-changed-as-a-result-umuc6vsde) · Security Boulevard · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#claude](https://daily.dev/tags/claude), [#salesforce](https://daily.dev/tags/salesforce), [#ai-governance](https://daily.dev/tags/ai-governance) [View this post on daily.dev](https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11) ```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":"Engineering Multi-Agent AI Teams That Build and Test Themselves","url":"https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11"},"datePublished":"2026-09-28T19:00:55.177Z","dateModified":"2026-09-28T19:01:21.782Z","description":"Salesforce's Marketing Cloud engineering team built Agent Designer, a system where an orchestrator and seven specialist AI agents design, verify, test, and...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/7ba21dbc4963fafab4d475c95a7e5b3f?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/7ba21dbc4963fafab4d475c95a7e5b3f?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Salesforce Engineering","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":"Salesforce Engineering","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/513375817ec24d97b9cdb5efd16f87f2","url":"https://daily.dev/sources/salesforceeng"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,claude,salesforce,ai-governance","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Salesforce Engineering","item":"https://daily.dev/sources/salesforceeng"},{"@type":"ListItem","position":3,"name":"Engineering Multi-Agent AI Teams That Build and Test Themselves"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/engineering-multi-agent-ai-teams-that-build-and-test-themselves-ehcixev11#faq","mainEntity":[{"@type":"Question","name":"How do you prevent AI agents from repeatedly trying to fix the wrong bug in a self-healing repair loop?","acceptedAnswer":{"@type":"Answer","text":"Bound the retry process with hard limits: Salesforce's Agent Designer restricts self-healing to two fix attempts and a $3 spending ceiling per team. A test doctor first classifies the failure type (violation, permission block, tool error, boot error, soft failure, infinite loop, or state corruption) before attempting repair, and the system escalates to a human instead of continuing to guess after limits are hit. Anyone building autonomous repair loops can compare bounded-retry strategies like this one on daily.dev."}},{"@type":"Question","name":"How do you keep AI verifier agents from either rubber-stamping bad designs or blocking good ones with false failures?","acceptedAnswer":{"@type":"Answer","text":"Require verifiers to output structured PASS, WARN, or FAIL results instead of free-form text, then run a meta-check comparing their outputs against each other and the proposed design. Rules like warning rather than failing when uncertain, and not failing a design solely for a specific permission flag, help distinguish real violations from reviewable concerns, since no single verifier is trusted in isolation. Teams designing agent verification pipelines can track patterns like this on daily.dev while refining their own checks."}},{"@type":"Question","name":"How should you calculate spending budgets for different multi-agent architectures like swarms versus single agents?","acceptedAnswer":{"@type":"Answer","text":"Apply a different formula per topology instead of one static limit: a flat check for a single agent, items multiplied by cost multiplied by steps for a swarm, and whole-roster ceilings for a team. Salesforce's Agent Designer stores these budgets, turn limits, and cost ceilings centrally, and also divides an orchestrator's own 500-turn budget across analysis, approval, generation, testing, and repair phases. Developers weighing budget controls across agent topologies can follow similar orchestration approaches on daily.dev."}}]} ```