--- title: "The NPCs Skipped the Script" newsletter: "User Community" date: 2026-10-15 source: https://aaif.live/newsletters/usercommunity/2026-10-15-the-npcs-skipped-the-script --- # The NPCs Skipped the Script *Plus... a 60% AI quota hit, reconstructing agent runs, and 30,000 daily merges.* *User Community — Agentic AI Foundation, 2026-10-15* I'll need the 43 days until Black Friday to set up my budget guardrails for this [https://www.theverge.com/news/1010365/shopify-opens-its-catalog-up-to-more-ai-agents]. ## Gems ## Reverse Engineering and the Future of Software ## CODING AGENTS LUNCH & LEARN What happens when developers crack open the tools they use and start building their own? Join us for an open discussion on reverse-engineering projects like REA [https://github.com/morluto/rea], what they've uncovered, and what they could mean for the future of software development. Bring your questions, share what you're working on, or just come along for the conversation. Friday, October 16 | 9–10 AM PT / 4–5 PM UTC JOIN LIVE [https://home.mlops.community/home/events/coding-agents-lunch-and-learn-session-28-reverse-engineering-and-the-future-of-software-4szbeg84t0] Find recordings of our previous lunch and learns here [https://www.youtube.com/playlist?list=PL3vkEKxWd-utz7ew8QmiuTFNjjkvvM_xM]! ## Inside a game where AI agents never sleep What happens when an MMORPG launches with thousands of AI agents already playing? Agent-populated worlds could create social dynamics from day one, while similar approaches are reshaping how games are built and tested. * Agent-driven gameplay: Replacing scripted NPCs with autonomous players could create evolving interactions, competition, and more persistent game worlds. * Multi-agent development: An internal framework coordinates frontend, backend, and QA agents through iterative loops, with humans focused on architecture and evaluating results. * Inference economics: Self-hosting open-weight models could deliver 80–90% of current performance at a tenth of the token cost, though GPU infrastructure and expertise complicate the calculation. The practical challenge is balancing agent autonomy with predictable outputs, operating costs, and human judgment. [https://podcasts.apple.com/gb/podcast/inside-a-game-where-ai-agents-never-sleep/id1505372978?i=1000793894520](https://podcasts.apple.com/gb/podcast/inside-a-game-where-ai-agents-never-sleep/id1505372978?i=1000793894520) [https://home.mlops.community/home/videos/inside-a-game-where-ai-agents-never-sleep](https://home.mlops.community/home/videos/inside-a-game-where-ai-agents-never-sleep) [https://open.spotify.com/episode/2ASSrxgdDdR8ONLHuBzQRD?si=_WdBHzvVTBWFDPJz84sVzA](https://open.spotify.com/episode/2ASSrxgdDdR8ONLHuBzQRD?si=_WdBHzvVTBWFDPJz84sVzA) ## Bridging the gap between AI builders and domain experts A background agent checking issues every five minutes burned through 60% of a weekly AI usage allowance. It raises an interesting question about whether narrowly scoped agents could handle repetitive tasks more efficiently than general-purpose models. * Specialized agents: Smaller models with tightly defined tools and optimized prompts could outperform larger, general-purpose setups on specific tasks. * Portable runtimes: Standardizing prompts, models, providers, and execution environments could make tested agents reproducible and shareable across systems. * Context overhead: Adding MCP servers and skills can increase context requirements and reduce effectiveness, making dedicated agents an attractive alternative. Reliable, reusable agents depend on matching each task with the right configuration and measurable performance. [https://podcasts.apple.com/gb/podcast/bridging-the-gap-between-ai-builders-and-domain-experts/id1505372978?i=1000795965497](https://podcasts.apple.com/gb/podcast/bridging-the-gap-between-ai-builders-and-domain-experts/id1505372978?i=1000795965497) [https://home.mlops.community/home/videos/bridging-the-gap-between-ai-builders-and-domain-experts](https://home.mlops.community/home/videos/bridging-the-gap-between-ai-builders-and-domain-experts) [https://open.spotify.com/episode/6Gq5bbfdQyWXivUqVB75df?si=nHRkWJHhRci_MiAh9onHzQ](https://open.spotify.com/episode/6Gq5bbfdQyWXivUqVB75df?si=nHRkWJHhRci_MiAh9onHzQ) ## I saw something in MCP. Then I spent 18 months building it. An agent changing network configurations needs more than a successful tool call to prove the network is safe. Eighteen months of MCP experiments led to NetClaw, an open-source network engineering agent designed around verifiable evidence, scoped permissions, and controlled execution. * Network evidence: MCP exposes pyATS and Genie observations, letting agents investigate device state and verify changes against actual network conditions. * Agent coordination: OpenClaw and Hermes runtimes support specialized agents working together, with explicit boundaries around capabilities and authorization. * Failure recovery: Request identifiers and recorded results help distinguish interrupted operations from failed ones, preventing unsafe automatic retries. The engineering challenge is making network automation inspectable, authorized, and recoverable when execution doesn't go as planned. [Read the blog](https://aaif.io/blog/i-saw-something-in-mcp-then-i-spent-18-months-building-it) ## Welcome to Glass Town: building a software factory with a paper trail An autonomous coding factory can burn through tokens before finishing the job, with failed runs adding to the bill. Glass Town explores how to make multi-agent software development repeatable, inspectable, and less expensive. * Persistent environments: Cloud-based Sprites isolate agent execution, preserve dependencies between sessions, and pause compute billing when idle. * Cost controls: Five-minute worker limits, cheaper models, disabled automatic retries, and scripts for non-LLM tasks help control spending. * Traceable changes: The Entire CLI links prompts, responses, and tool calls to Git commits, while AGENTS.md [http://AGENTS.md] instructions enforce per-file checkpoints. The architecture shows how persistent execution, bounded agent workloads, and commit-linked histories can make autonomous development easier to reproduce and audit. [Read the blog](https://aaif.io/blog/welcome-to-glass-town-building-a-software-factory-with-a-paper-trail) ## Agent identity and delegated access in MCP systems One GitHub update can cross several agents, MCP servers, and downstream services, each making its own identity and authorization decision. The challenge is preserving who initiated the work, who is acting now, and exactly which permissions should survive each hop. * User-delegated OAuth, workload identities, and RFC 8693 token exchange cover different authorization patterns without collapsing everything into one shared credential. * Delegated agents should receive narrower, task-specific authority, with reauthorization where long-running tasks outlive approvals or permissions. * OpenTelemetry, W3C Trace Context, and MCP trace propagation help reconstruct execution alongside authorization records. Reliable agent access depends on keeping identity, delegated authority, and execution history connected across the full request path. [Read the blog](https://aaif.io/blog/agent-identity-and-delegated-access-in-mcp-systems) ## Can you reproduce an agent run? An agent approves a refund after checking account records, but six months later, nobody can reconstruct why. Reproducing agent behavior requires more than saving prompts and code versions, particularly when tools, memory, permissions, and external data can change. * Reconstruction vs. replay: Recorded execution histories explain past decisions, while fresh reruns may produce different results as models and external systems change. * Execution evidence: Model configurations, retrieved content, tool responses, state checkpoints, and human approvals help establish what influenced each action. * Retention tradeoffs: OpenTelemetry can capture detailed execution data, but sensitive content requires selective logging, access controls, and retention limits. Reliable investigations depend on preserving the original execution conditions rather than expecting agents to behave identically twice. [Read the blog](https://aaif.io/blog/can-you-reproduce-an-agent-run) ## What’s happening across the chapters ## LOCAL ORGANIZERS Huge shout-out to Ijeoma Onwuka [https://www.linkedin.com/in/onwuka-ijeoma/] for all her hard work coordinating with our community organizers and helping bring these events together. She's also put together a great roundup of last week's meetups, covering Singapore, Hyderabad, Kolkata, and Pune, with a look at what the communities have been up to and the local organizers making it happen on the ground. Check out her post for the full roundup. [https://lnkd.in/p/eBDdhWRk] ## IN-PERSON EVENTS ## Come and connect * London [https://luma.com/9prdte0x] - October 15 * Berlin [https://luma.com/3q4cjah8] - October 15 * Bengaluru [https://luma.com/adhicxnt] - October 17 * Silicon Valley [https://luma.com/8dzsovxr] - October 21 * Luxembourg [https://luma.com/clx51oyb] - October 21 * Auckland [https://luma.com/rhsbtg3i] - October 21 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 * Accuracy & Reliability of AI Agents [https://home.mlops.community/home/events/accuracy-and-reliability-of-ai-agents-virtual-summit-77nje0s530] - October 15 * AAIF Reading Group [https://home.mlops.community/home/events/coding-agents-lunch-and-learn-session-28-reverse-engineering-and-the-future-of-software-4szbeg84t0] - October 16 * Live from AGNTCon + MCPCon North America [https://luma.com/aaif-n7w0] - October 22 --- Source: https://aaif.live/newsletters/usercommunity/2026-10-15-the-npcs-skipped-the-script