Commit & Push

AI Agents, RAG, and the Gap Between Hype and Execution

Host Damien Filiatrault talks with Micah Johnson, co-founder of Biggest Goal, for a practical conversation about what AI agents really are, where they’re genuinely useful today, and why so many companies still struggle to turn AI enthusiasm into real business results. Drawing on his experience building agents and training teams across leadership, operations, and management, Micah breaks down the building blocks of agentic workflows in plain language: instructions, tools, triggers, loops, RAG databases, and the systems that make AI useful beyond the hype. The conversation moves from definition to execution. Micah explains how agents differ from normal chat-based AI, why they work especially well for operational use cases like analysis and internal workflows, and how tools like N8N and Raggy have made it dramatically easier for teams to build useful automations without heavy engineering overhead. Damien and Micah also dig into the tradeoffs: when AI is actually necessary, when traditional automation might be enough, and why giving agents too many tools or too much context can make them less reliable instead of more capable.  Just as importantly, the episode explores why so many AI projects fail. According to Micah, the issue usually is not the model itself. It is the lack of structure, standardization, training, and strategic thinking around how teams adopt these tools. Instead of assuming ChatGPT, Copilot, or Claude will magically fix broken workflows, companies need better systems, clearer use cases, and stronger leadership alignment if they want AI to create lasting value. Tune in for a grounded, jargon-light tour of agents, RAG, N8N, no-code automation, and what it really takes to move from experimentation to execution with AI. What you’ll learn * What an AI agent actually is, and how it differs from a standard chatbot * How agents can be triggered by events inside tools like Monday or ClickUp and take action automatically * Why data analysis and internal knowledge retrieval are two of the strongest real-world use cases today * How RAG systems turn folders of SOPs and documents into searchable, AI-friendly knowledge bases * Why tools like Raggy and N8N make it possible to build useful agent workflows quickly, even without deep engineering work * Why narrow, focused agents tend to perform better than overloaded ones with too many tools or too much context * What makes N8N stand out from tools like Zapier and Make for agentic workflows * Why so many AI projects fail inside companies, even when the tools themselves are powerful * Why AI adoption needs process design, training, and leadership alignment—not just subscriptions and enthusiasm Memorable sound bites * “Agents will look at their instructions, look at their tools, do something, and then circle back to themselves.” * “Fill in those gaps in your business all day long with tiny little simple agents.” * “Keep it as narrow of a focus for an agent as possible.” * “You literally just change the prompt in plain language. You’re not reconstructing anything.” * “You’re just throwing a tool at a problem.” * “How do we get them past just the individual small gains and actually building standardized systems?” Get 20% off your first month with Scalable Path: https://www.scalablepath.com/commit ‍ [https://www.scalablepath.com/commit]Commit & Push Website: https://www.commit-push.com/ ‍ [https://www.commit-push.com/]Scalable Path Website: https://www.scalablepath.com/ [https://www.scalablepath.com/]

Zero-Click Commerce and the Rise of Buying Agents: A conversation with Kevin Williams

Host Damien Filiatrault sits down with Kevin Williams, founder of Ascend AI, to map the fast-shifting terrain of agentic commerce—the move from “LLMs that recommend” to systems that can discover, compare, and eventually complete purchases with minimal human clicking. Drawing on his background building direct-to-consumer brands and solving post–iOS 14 attribution gaps with machine learning, Kevin explains why “zero-click commerce” is coming, why brands are already seeing meaningful drops in organic traffic, and what they can do to stay discoverable as shopping flows migrate into ChatGPT/Gemini-style interfaces. Kevin breaks the space into two big layers: being found (AEO/GEO) and being transactable (a growing soup of protocols, feeds, and enrichment standards). He walks through how Google’s ecosystem is evolving via Merchant Center feed enrichment, why OpenAI and Google are unlikely to converge quickly on a single standard, and why most non-Shopify brands may need to support multiple feeds/endpoints in the near term. The conversation also digs into what this shift means for marketers: fewer familiar on-site analytics signals, more “share of voice” style measurement, and a future where prompt-level attribution becomes the new battleground. What you’ll learn * What “agentic commerce” means today vs. what it implies next: LLM-assisted discovery now, agent-assisted purchasing later (especially in B2B replenishment and procurement workflows). * Why zero-click commerce changes the operational burden for merchants: identity, payment security, and fulfillment still have to happen—just without a traditional checkout flow. * The difference between AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization), and why GEO increasingly depends on broader web presence—reviews, sentiment, and third-party references—not just on-site content. * Why the web-as-screenshots approach (agentic browsers) is clunky and expensive, and why platforms are pushing toward structured data feeds and “traction points” that models can reliably consume. * How Google’s approach leans on Merchant Center product feeds with new enrichment fields, and why populating those fields (sentiment/occasion/recipient) becomes a real scaling challenge for large catalogs. * Why there’s no “one feed to rule them all” yet—and why competing incentives (ads, attribution, control) make standardization hard in the short term. * What this shift does to analytics: less time-on-site and scroll depth, more reliance on referral signals, model “share of voice” tracking, and paid-placement pressure. * Who adopts fastest: brands closest to the traffic cliff (digitally native, younger-skewing audiences) vs. segments that haven’t felt the drop as sharply. Memorable sound bites * “Zero-click commerce means the entire transaction happens in the LLM.” * “We have to clarify the matrix for the model so it’s not guessing from screenshots.” * “AEO is relatively straightforward. GEO is the bigger lift.” * “There is no one feed to rule them all yet.” * “It’s going to be really rough” (for traditional analytics in a world where fewer humans visit your site). Tune in for a practical, acronym-heavy (but real-world grounded) tour of what commerce looks like when discovery and conversion start moving inside LLMs—and what builders and brands can do now to avoid getting buried as the interfaces change. Get 20% off your first month with Scalable Path: https://www.scalablepath.com/commit [https://www.scalablepath.com/commit] Commit & Push Website: https://www.commit-push.com/ [https://www.commit-push.com/] Scalable Path Website: https://www.scalablepath.com/ [https://www.scalablepath.com/]

Collaborating with AI on Infrastructure, Safely: A conversation with Brit Myers

Host Damien Filiatrault welcomes Brit Myers, VP of Engineering at System Initiative, to unpack a new model for infrastructure automation built for the realities Terraform and traditional GitOps often struggle with: drift, legacy systems, manual console “stop the bleed” fixes, and the rising need to collaborate safely with AI. Brit explains why System Initiative is betting on a “digital twin” of your infrastructure—a one-to-one model that holds intent, current reality, and the transitions between them—so teams get faster feedback, tighter control over change, and a lot less “merge and pray.” What you’ll learn * What System Initiative is building: an AI-native infrastructure automation platform that can handle the same categories of work as Terraform and GitOps pipelines, but with different core assumptions. * The first assumption they challenge: infrastructure automation doesn’t have to be only static, declarative files that get versioned and processed later. * The second assumption: state and reconciliation shouldn’t live as a separate, fragile artifact (like a state file) that engineers sometimes have to manually repair. * Why “all changes must go through Terraform” breaks down in the real world—especially during incidents, cost overruns, or when someone has to make an emergency console change without creating future landmines. * Where the pain is worst: organizations with long-lived, legacy, or stateful systems (common in insurance and financial services) that can’t cleanly modernize everything at once. * How System Initiative uses programmable TypeScript functions to define behavior: propagation of shared values (like regions), organizational policy checks, compliance guardrails, and security logic. * How “change sets” work: safe simulations of proposed infrastructure updates that show not just what you want to change, but how it would ripple through the system and what actions (create/update/delete) would be required in reality. * Why this is especially useful with AI agents: you can bring an external agent (Claude, etc.) to propose changes through an API, iterate inside a change set, and only apply when a human—or your checks—approves. * A concrete AI workflow: using an architecture diagram as input, an agent can iteratively build a high-fidelity model and propose infrastructure in minutes while you watch changes happen live in the collaborative UI. * What’s live today: a multi-tenant SaaS at Systeminit.com with AWS coverage now, and plans to expand to GCP, Azure, DigitalOcean, and more—plus the ability for users to author and extend schemas/functions inside the product. Memorable sound bites * “We want a little less ‘merge and pray’ and a little more ‘yeah, I got this.’” * “Terraform assumes all changes are made through Terraform—and that’s not how the world works.” * “You should be able to stop the bleed in the console without creating baggage later.” * “Everything is a function—behavior, policy, even what an asset is—and it’s all TypeScript.” * “Change sets let you collaborate with AI in safety, before anything touches production.” * “We haven’t found an AI use case that flops yet—bring us the one you think will.” Tune in for a practical look at infrastructure automation after drift: digital twins, simulation-first changes, and how to let AI help with ops work without handing it the keys to production. System Initiative Discord: https://discord.gg/system-init [https://discord.gg/system-init] Get 20% off your first month with Scalable Path: https://www.scalablepath.com/commit [https://www.scalablepath.com/commit] Commit & Push Website: https://www.commit-push.com/ [https://www.commit-push.com/] Scalable Path Website: https://www.scalablepath.com/ [https://www.scalablepath.com/]

Building Smarter Dev Environments for Humans and AI: A conversation with Rob Whiteley

Host Damien Filiatrault welcomes Rob Whiteley, CEO of Coder, for a grounded tour of how AI agents are reshaping software development, from cloud-based dev environments to “software-intern” agents that can refactor codebases for hours at a time. They dig into why infrastructure and context matter more than model choice, how Anthropic runs Claude Code as a first-class “developer,” and what it really takes for startups and enterprises to trust agents with real work. What you’ll learn * How Coder turns your laptop-centric workflow into a centralized, cloud-based development platform that provisions compute, GPUs, tools, and credentials as code. * Why code completion is no longer the “end game,” and how developers are moving from line-by-line autocomplete to truly agentic workflows and background tasks. * How Anthropic runs Claude Code in a walled-off workspace (with its own tools, Terraform-defined context, and MCP-powered toolbelt) and why that pattern points to the enterprise future. * The two essentials for productive agents: solid infrastructure (VMs/containers, GPUs, access to Git, browsers, etc.) and rich, structured context about their environment. * System prompts vs. user prompts: how hidden “agent personalities” work under the hood, and why conflicting instructions can quietly tank an agent’s effective IQ. * Practical patterns for startups vs. big companies: cursor + Coder for smaller teams, and Bedrock-backed stacks (Q, Cursor, Claude Code) for enterprises that need governance and data control. * Why agent adoption follows a “bathtub curve”, junior and principal engineers love them, mid-levels are skeptical, and how to design prompts, tools, and workflows that flatten that curve. * A realistic roadmap to long-running agents: when it makes sense to let a model refactor codebases or decouple a front end from its backend over hours instead of minutes. * Why “English is the new programming language,” what that means for vibe coders and systems thinkers, and how non-engineers are becoming their team’s internal app builders. * How to think about agents like summer interns: what it takes to train them, where they shine, and why your culture around mentoring junior talent predicts your AI success. Memorable sound bites * “Agents are just a gen-AI call in a loop—what matters is the tools and context you give that loop.” * “Most people deployed naked agents, starved them of tools, and then decided agents ‘aren’t ready.’” * “Claude Code reads its own Terraform file on boot. It literally learns who it is and where it’s running.” * “If you’d never hire summer interns because they’re ‘too much work,’ you’re going to hate agents.” * “A developer isn’t just a coder anymore—they’re a systems thinker who can break problems down and speak clearly in plain English.” * “We may end up with fewer traditional software engineers—but many more developers building software.” Tune in for a candid, tactical look at AI-native development: how to provision agent workspaces, avoid trust-killing misconfigurations, and turn agents from novelty toys into reliable collaborators for both startups and large engineering orgs. Get 20% off your first month with Scalable Path: https://www.scalablepath.com/commit [https://www.scalablepath.com/commit] Commit & Push Website: https://www.commit-push.com/ [https://www.commit-push.com/] Scalable Path Website: https://www.scalablepath.com/ [https://www.scalablepath.com/]

Beyond the Happy Path: Josh Clark & Veronika Kindred on “Sentient Design”

Host Damien Filiatrault welcomes Josh Clark and Veronika Kindred—principal and researcher at Big Medium and co-authors of the forthcoming book Sentient Design—for a cross-generational look at how AI changes product and UX. They unpack what “sentient” really means in interfaces, why defensive design matters more than ever, and how to let intelligence co-pilot the presentation layer without giving up control. What you’ll learn * A practical definition of sentient design: context-aware, radically adaptive, collaborative, deferential, and ambient interfaces—without implying machine consciousness. * How Gen Z “AI-native” instincts and decades of UX experience productively clash to re-question sacred “best practices.” * Patterns for bespoke UI: letting AI assemble trusted design-system components (e.g., Salesforce’s generative canvas) to meet intent in the moment. * The NPC pattern: AI as a visible participant in multi-user tools (comments, cursors, suggestions) instead of a side-chat. * Defensive design: guardrails, explainability, reversible actions, and graceful degradation—designing for failure paths when there’s no single “happy path.” * Why LLMs excel at intent translation (driving presentation choices) but shouldn’t be your source of truth. Memorable sound bites * “We can create genuinely new experiences when we weave intelligence into the interface.” * “There is no happy path anymore—so design guardrails, not just flows.” * “Delegate decisions, don’t abdicate them—suggest, don’t impose.” * “LLMs aren’t answer machines; they’re great at interpreting intent.” * “When the escalator fails, it turns into stairs—your product should, too.” Tune in for grounded tactics to add AI as a design material: smarter dashboards, safer autonomy, and collaboration patterns that keep users in control while moving faster. Get 20% off your first month with Scalable Path: https://www.scalablepath.com/commit [https://www.scalablepath.com/commit] Commit & Push Website: https://www.commit-push.com/ [https://www.commit-push.com/] Scalable Path Website: https://www.scalablepath.com/ [https://www.scalablepath.com/]