--- title: "Less config, fewer confused agents" newsletter: "User Community" date: 2026-09-03 source: https://aaif.live/newsletters/usercommunity/2026-09-03-less-config-fewer-confused-agents --- # Less config, fewer confused agents *Plus: stronger tool contracts, A2A changes, and where coding agents break.* *User Community — Agentic AI Foundation, 2026-09-03* I love seeing how much is happening across the AAIF community. Sunnyvale [https://www.linkedin.com/posts/vijaybhore_artificialintelligence-agenticai-llms-ugcPost-7499168450242809856-BDpi/] heard talks on memory architectures and state persistence for agents, while Pittsburgh [https://www.linkedin.com/posts/jonzeolla_our-first-agentic-ai-foundation-pittsburgh-share-7499097787217739777-C8ry/] went hands-on building and testing an MCP server from scratch. There’s plenty more on the calendar too, with local meetups and virtual sessions coming up. Check out the events section to see what’s happening near you, or what you can join online. ## Gems ## Where are coding agents breaking? ## LIVE + INTERACTIVE - LUNCH AND LEARN What can the Hugging Face security incident teach us about the risks emerging around coding agents? Ads Dawson from Dreadnode joins this week’s Lunch & Learn to unpack what happened, the vulnerabilities builders should be paying attention to, and what he’s seeing across agent security right now. Then it opens up to the room. Bring a question, share something you’ve run into, or just listen in as the conversation goes wherever the useful stuff is. Friday, September 4 · 9 AM PT / 6 PM CEST REGISTER NOW [https://home.mlops.community/home/events/coding-agents-lunch-and-learn-session-23-whats-new-in-coding-agents-ig8sug5bk5?agenda_day=6a95be6212d9e72e119b6164&agenda_track=6a95be6312d9e72e119b6179&agenda_stage=6a95be6212d9e72e119b6169&agenda_filter_view=stage&agenda_view=list] Can’t make that time? We’re expanding Lunch & Learns across US, EU, and APAC-friendly sessions, so keep an eye out for more dates and times. ## The MLOps Adventure Continues: An AGENTS.md Ready Stack for AI/ML Removing 900+ lines of config while doubling static checks sounds backwards, but it came from giving AI coding agents fewer places to get confused. The redesign centers on deterministic repository feedback: one canonical gate, executable guardrails, and isolated test state. * One gate: mise run all now drives local workflows, git hooks, CI, and agent instructions, removing duplicated command lists that had silently drifted. * Fast isolation: copying a pre-migrated SQLite template cut MLflow test overhead by about 60% after migrations had pushed runtime from 34 to 339 seconds. * Executable rules: AGENTS.md guides behavior, while tests, linters, and dependency checks verify that commands, paths, and security overrides still hold. The core lesson is that agents work best when repository rules produce one clear, repeatable source of truth. [Read the blog](https://mlops.community/blog/the-mlops-adventure-continues-an-agents-md-ready-stack-for-ai-ml) ## How predictive analytics stops budget overruns before they happen A model change can push token usage up 8x, while retry-heavy agent loops can burn through budget before conventional billing alerts catch the problem. The discussion looks at how FinOps can move closer to the engineering workflow as AI usage scales. * Predictive budgets can flag teams or service accounts heading far beyond quota early in the month, giving engineers time to adjust rather than simply hitting a hard stop. * FinOps checks can be built into system prompts, pull requests, and deployment workflows, surfacing policy, model, and infrastructure costs as code changes. * Some workflows may be better served by rules, retrieval, or conventional ML, with LLMs reserved for the steps that genuinely need generation. The core idea is to make cost an engineering signal that appears early enough to shape architecture, model choice, and day-to-day decisions. [https://podcasts.apple.com/us/podcast/how-predictive-analytics-stops-budget-overruns-before/id1505372978?i=1000784610595](https://podcasts.apple.com/us/podcast/how-predictive-analytics-stops-budget-overruns-before/id1505372978?i=1000784610595) [https://home.mlops.community/home/videos/how-predictive-analytics-stops-budget-overruns-before-they-happen](https://home.mlops.community/home/videos/how-predictive-analytics-stops-budget-overruns-before-they-happen) [https://open.spotify.com/episode/3A20FeJFBHIsLGESe3imYX?si=R1r7I17LQcSfjW4qOHFoOw](https://open.spotify.com/episode/3A20FeJFBHIsLGESe3imYX?si=R1r7I17LQcSfjW4qOHFoOw) ## The Winchester Mystery House Problem in AI development A workflow that cost $1 per 1,000 records at 90% accuracy reached 95% accuracy while costing less, after an optimizer moved 75% of the work out of the model and into code. That result sits inside a broader discussion about when agentic behavior should become a defined workflow. * Models increasingly inherit assumptions from the coding harnesses they were trained around, which can make custom harnesses harder to build. * Repeated agent tasks can often be “crystallized” into smaller models or conventional code once the process is understood. * DSPy separates task definitions from implementation, allowing prompts, models, and even harness code to be optimized without redefining the task. The practical direction is toward identifying which tasks still need open-ended agents and which are mature enough to become cheaper, more reliable workflows. [https://podcasts.apple.com/us/podcast/the-winchester-mystery-house-problem-in-ai-development/id1505372978?i=1000785551986](https://podcasts.apple.com/us/podcast/the-winchester-mystery-house-problem-in-ai-development/id1505372978?i=1000785551986) [https://home.mlops.community/home/videos/the-winchester-mystery-house-problem-in-ai-development](https://home.mlops.community/home/videos/the-winchester-mystery-house-problem-in-ai-development) [https://open.spotify.com/episode/4IjjWOjSq8XQPAXLq6ApRQ?si=TnlufikXRKqxS8ZHSpyYCQ](https://open.spotify.com/episode/4IjjWOjSq8XQPAXLq6ApRQ?si=TnlufikXRKqxS8ZHSpyYCQ) ## Improving tool-call reliability with JSON Schema 2020-12 A $12,500 transfer missing one approval field can either trigger an opaque failure or a precise, repairable error before the ledger is touched. JSON Schema 2020-12 gives MCP tools a stronger way to encode those rules directly into the contract rather than leaving them in descriptions. * Conditional validation: if/then, oneOf, and $defs can enforce state-dependent requirements and prevent incompatible argument combinations. * Clearer recovery: schema failures can return field-specific invalid-params errors that an orchestrator repairs within a bounded retry loop. * Layered enforcement: validation happens before business logic, while the underlying service still keeps final authority over business rules. The key design shift is treating tool schemas as executable API contracts that tell agents what can be fixed and what should stop. [Read the blog](https://aaif.io/blog/improving-tool-call-reliability-with-json-schema-2020-12) ## A2A v1.0: a builder's guide, part 2 - migration, security, and production A2A v1.0 changed enough on the wire that a 0.3 parser can break even when the task flow looks familiar. The migration also introduces important changes around bindings, durable tasks, agent-card verification, and authorization. * Breaking changes: kind discriminators are gone, enums changed, and ProtoJSON is now the canonical serialization. * Durability: ListTasks, database-backed stores, and persistent webhooks help work survive lost IDs, disconnects, and restarts. * Security: signed agent cards protect endpoint integrity, while AUTH_REQUIRED supports controlled approval and credential escalation. The practical takeaway is how to migrate A2A systems while preserving compatibility, security boundaries, and durable task behavior. [Read the blog](https://aaif.io/blog/a2a-v1-0-a-builder-s-guide-part-2-migration-security-and-production) ## The MLOps adventure continues: An AGENTS.md ready stack for AI/ML Removing 900+ lines of config while doubling static checks sounds backwards, but it came from giving AI coding agents fewer places to get confused. The redesign centers on deterministic repository feedback: one canonical gate, executable guardrails, and isolated test state. * One gate: mise run all now drives local workflows, git hooks, CI, and agent instructions, removing duplicated command lists that had silently drifted. * Fast isolation: copying a pre-migrated SQLite template cut MLflow test overhead by about 60% after migrations had pushed runtime from 34 to 339 seconds. * Executable rules: AGENTS.md guides behavior, while tests, linters, and dependency checks verify that commands, paths, and security overrides still hold. The core lesson is that agents work best when repository rules produce one clear, repeatable source of truth. [Read the blog](https://mlops.community/blog/the-mlops-adventure-continues-an-agents-md-ready-stack-for-ai-ml) ## Improving tool-call reliability with JSON Schema 2020-12 A $12,500 transfer missing one approval field can either trigger an opaque failure or a precise, repairable error before the ledger is touched. JSON Schema 2020-12 gives MCP tools a stronger way to encode those rules directly into the contract rather than leaving them in descriptions. * Conditional validation: if/then, oneOf, and $defs can enforce state-dependent requirements and prevent incompatible argument combinations. * Clearer recovery: schema failures can return field-specific invalid-params errors that an orchestrator repairs within a bounded retry loop. * Layered enforcement: validation happens before business logic, while the underlying service still keeps final authority over business rules. The key design shift is treating tool schemas as executable API contracts that tell agents what can be fixed and what should stop. [Read the blog](https://aaif.io/blog/improving-tool-call-reliability-with-json-schema-2020-12) ## A2A v1.0: a builder's guide, part 2 - migration, security, and production A2A v1.0 changed enough on the wire that a 0.3 parser can break even when the task flow looks familiar. The migration also introduces important changes around bindings, durable tasks, agent-card verification, and authorization. * Breaking changes: kind discriminators are gone, enums changed, and ProtoJSON is now the canonical serialization. * Durability: ListTasks, database-backed stores, and persistent webhooks help work survive lost IDs, disconnects, and restarts. * Security: signed agent cards protect endpoint integrity, while AUTH_REQUIRED supports controlled approval and credential escalation. The practical takeaway is how to migrate A2A systems while preserving compatibility, security boundaries, and durable task behavior. [Read the blog](https://aaif.io/blog/a2a-v1-0-a-builder-s-guide-part-2-migration-security-and-production) ## IN-PERSON EVENTS ## Come and connect * Pittsburgh [https://luma.com/eob7xzdb] - September 3 * Shanghai [https://luma.com/m4ygi0b2] - September 5 * AGNTCon + MCPCon China [https://www.lfopensource.cn/mcp-dev-summit-shanghai/] - September 6 * Atlanta [https://luma.com/t2ysyedz] - September 8 * Seoul [https://luma.com/sa692amj] - September 9 * Toronto [https://luma.com/p6vg7zag] - September 10 * AGNTCon + MCPCon Japan [https://events.linuxfoundation.org/agntcon-mcpcon-japan/] - September 10 * Pune, India [https://luma.com/4h0mjfzr] - September 12 * New York [https://luma.com/aaif-kc6u] - September 15 * AGNTCon + MCPCon Europe [https://events.linuxfoundation.org/agntcon-mcpcon-europe/] - September 17 Find your city here [https://aaif.io/events?tab=community], or start a chapter if there isn't one yet. ## VIRTUAL EVENTS ## Join from anywhere * Reading Group: Prompt Injection as Role Confusion [https://home.mlops.community/home/events/prompt-injection-as-role-confusion-rethinking-agent-security-kzdnmrqqqz?agenda_day=6a7a2823abc455f7962051e5&agenda_track=6a7a2823abc455f7962051fc&agenda_stage=6a7a2823abc455f7962051eb&agenda_filter_view=stage&agenda_view=list] - September 3 * Lunch and Learn - Session 23 [https://home.mlops.community/home/events/coding-agents-lunch-and-learn-session-23-whats-new-in-coding-agents-ig8sug5bk5?agenda_day=6a95be6212d9e72e119b6164&agenda_track=6a95be6312d9e72e119b6179&agenda_stage=6a95be6212d9e72e119b6169&agenda_filter_view=stage&agenda_view=list] - September 4 * 36 Hours Live from AGNTCon + MCPCon Europe [https://luma.com/aaif-afd7] - September 17 - 18 --- Source: https://aaif.live/newsletters/usercommunity/2026-09-03-less-config-fewer-confused-agents