OpenAI's Dev Day dropped some major announcements this month, but is AgentKit really revolutionary or just another "me too!" moment? The AI, Actually crew shares their reactions to the latest OpenAI releases, and digs into how to successfully implement AI agents in the real world.

In this episode, Jim Johnson steps in as host alongside Mike Finley, with special guests Nicole Kosky (who leads AnswerRocket's AI Business Transformation Practice) and Reilly Carrolll (Senior AI Solutions Consultant). Together, they tackle the practical, nitty-gritty challenges of bringing agents to life for enterprise clients—from gathering requirements that users don't know they have, to managing the surprising differences between what stakeholders say they need versus what they actually ask once they're hands-on with an agent.

Topics covered:

• OpenAI's Dev Day announcements and what they actually mean for enterprises
• Why successful agent implementations should follow the Software Development Lifecycle (SDLC)
• The critical role of context in agent performance
• Why user questions change dramatically from requirements phase to hands-on testing
• The emerging discipline of "agentic operations" and why it's non-negotiable
• Starting small: the power of quick wins over trying to boil the ocean

Follow the Gang:

Jim Johnson, AnswerRocket, Managing Partner - https://www.linkedin.com/in/jim-johnson-bb82451/
Mike Finley, AnswerRocket, CTO - https://www.linkedin.com/in/mikefinley/ 

Reilly Carroll, AnswerRocket, Senior AI Solutions Consultant - https://www.linkedin.com/in/reilly-carroll/ 

Nicole Kosky, AnswerRocket, Senior Director of Services - https://www.linkedin.com/in/nicole-kosky-5b9a3b6/ 

Chapters:

00:00      Introduction and Guest Introductions

01:30      OpenAI Dev Day Announcements

05:25      Understanding AI Agents

09:05      Practical Implementation of AI Agents

11:11      Challenges in Client Engagement

14:17      Agent Development and User Experience

16:55      The Challenge of Capturing Agent Requirements

19:52      The Importance of Agentic Operations

22:40      Navigating the Future of AI Agents

29:09      Final Thoughts and Advice

Keywords: AI agents, OpenAI Dev Day, agent development, enterprise AI implementation, agentic operations, software development lifecycle, AI business transformation, agent context, LLM applications, practical AI strategy

Tired of AI agents that forget context mid-conversation or drift subtly off course in production? You're not alone. In this episode, the AI, Actually crew unpacks six critical engineering principles for building reliable AI agents—principles that separate proof-of-concepts from production-ready systems.

Pete, Mike, Andy, and Stew break down insights from AI expert Nate B. Jones, translating technical concepts into business-focused guidance. They explore why AI memory isn't just about storage, how to bound uncertainty without killing creativity, and why monitoring AI systems requires a completely different approach than traditional software.

This episode covers:

• Why stateful intelligence and memory management are fundamental to useful AI interactions
• How to engineer controls that bound uncertainty without over-constraining your models
• The shift from binary failures to subtle quality drift in AI systems
• Capability-based routing: matching the right model to the right job
• Post-production monitoring strategies that catch problems before your users do
• Continuous validation techniques for multi-turn agent conversations

This episode of AI, Actually centers around a video by @nate.b.jones about the 6 principles of AI Agents. That video can be watched in its entirety here: I've Built Over 100 AI Agents: Only 1% of Builders Know These 6 Principles

Follow the Gang:

• Mike Finley, CTO, AnswerRocket - https://www.linkedin.com/in/mikefinley/ 
• Pete Reilly, COO, AnswerRocket - https://www.linkedin.com/in/petereilly 
• Andy Sweet, VP Enterprise AI Solutions, AnswerRocket - https://www.linkedin.com/in/andrewdsweet/ 
• Stew Chisam, Operating Partner, StellarIQ - https://www.linkedin.com/in/stewart-chisam-7242543/ 

Chapters: 

00:00    Introduction to AI Agents and Engineering Principles

01:34     Introducing Nate B. Jones' AI Engineering Principles

03:03    Stateful Intelligence

10:16     Bounded Uncertainty

19:55     Intelligent Failure Detection

20:51     Evaluating LLM Responses

22:16     Monitoring Quality and Performance

23:53    Active Maintenance of LLM Systems

26:18     Understanding Subtle Failures

26:55    Capability-Based Routing

30:22    Aligning Models with Business Processes

33:41     Nuanced Health State Monitoring

37:36     Continuous Input Validation

41:36     Closing Thoughts

Keywords: AI agents, agentic AI, AI engineering, AI memory, stateful intelligence, AI monitoring, capability-based routing, AI evaluation, production AI, enterprise AI, AI agent development, LLM engineering, AI testing, AI agent failures, AI system monitoring

Sequoia says AI is a $10 trillion opportunity. But how do you actually capture it? In this episode, the AI, Actually crew tackles the gap between AI's promise and its practical deployment in the enterprise. From bold predictions about agent automation to Palantir's forward deployed engineer model, we explore what it really takes to move beyond ChatGPT licenses to actual business transformation.

The discussion gets real about the current state of AI agents—including why they'd rather fake your data than admit they're stuck—and examines the critical role of specialization in making AI work for specific business processes. Whether you're trying to onboard suppliers across multiple systems or wondering why your coding agent just hardcoded the test answers, this episode provides the unfiltered truth about where enterprise AI stands today.

Key topics covered:

• Sequoia's $10 trillion services automation thesis and historical parallels to the steam engine
• The forward deployed engineer model and why proximity to the problem matters
• Current state of AI agents: what's working, what's broken, and why they love mocking up data
• The critical importance of specialization vs. general-purpose AI
• Real enterprise integration challenges beyond Google Drive
• Replit's Agent 3 and the evolution of autonomous coding tools

Follow the Gang:

• Alon Goren, Founder & CEO, AnswerRocket - https://www.linkedin.com/in/alon-goren-87889681/
• Pete Reilly, COO, AnswerRocket - https://www.linkedin.com/in/petereilly
• Shanti Greene, Head of Data Science and AI Innovation, AnswerRocket - https://www.linkedin.com/in/shantigreene
• Stew Chisam, Operating Partner, StellarIQ - https://www.linkedin.com/in/stewart-chisam-7242543/

Chapters:

00:00 Introduction

01:14 Sequoia's $10 Trillion AI AI Thesis

08:45 The Specialization Problem: From General Purpose to Actually Useful

16:56 Forward Deployed Engineers: Digging into the Palantir Model

22:47 The Intelligence Revolution vs. The Information Revolution

24:02 Prototyping and User Engagement

26:07 The Role of Business Analysts in AI Deployment

29:24 The Year of the Agent Reality Check: Replit and Autonomous Coding

35:09 Mock Data and Unit Test Cheats: Watch Out for These AI Coding Traps

#aiactually #enterpriseaiadoption #forwarddeployedengineers #palantirFDEmodel #AISpecialization #businessprocessautomation #ReplitAgent3 #AIScaffolding #EnterpriseIntegration #aiROI

The AI Actually crew tackles the pressing concerns keeping enterprise leaders up at night: shadow AI infiltrating organizations, the crucial distinction between machine learning and LLMs, and why context engineering matters more than prompt engineering. Jim Johnson takes the moderator chair, joined by regular Mike Finley and special guests Andy Sweet (Advanced Models Practice Lead) and Shanti Greene (Head of Data Science and AI Innovation).

Topics covered:

• Shadow AI: Why employees bypass IT to use personal AI tools
• Machine Learning vs LLMs: Understanding when each technology wins
• Prompt vs Context Engineering: Moving beyond moving commas
• Local vs cloud models: When ownership makes sense (and when it doesn't)
• Agent operations and the challenge of model stability
• The surprising costs of AI tokens, especially for audio applications

Follow the Gang

Jim Johnson, AnswerRocket, Managing Partner - https://www.linkedin.com/in/jim-johnson-bb82451/
Mike Finley, AnswerRocket, CTO - https://www.linkedin.com/in/mikefinley/ 

Andy Sweet, AnswerRocket, VP Enterprise AI Solutions - https://www.linkedin.com/in/andrewdsweet/
Shanti Greene, AnswerRocket, Head of Data Science and AI Innovation - https://www.linkedin.com/in/shantigreene/ 

Chapters:

00:00 Introduction to AI Actually and the Team

01:59 Kimi Model Release: Should We Care?

06:59 Shadow AI Definition and Enterprise Impact

12:08 Leveraging Machine Learning and LLMs Together

26:00 Prompt Engineering vs. Context Engineering

28:13 Using LLMs to Write Prompts

29:56 Memory and Agent Ops

33:05 AI Literacy and Context Engineering

34:27 Stability and Model Changes

37:28 The Harvard MBA Analogy for AI Agents

43:01 Local Open Source AI Models: Pros & Cons

49:23 The Real Costs of Enterprise AI

Keywords: shadow AI, machine learning vs LLMs, context engineering, prompt engineering, enterprise AI, local models, agent operations, AI costs, Kimi model, AI governance

The MIT study claiming 95% of AI pilots fail has everyone talking—but what's really behind these failures? In this episode, the AI, Actually crew tackles the hard truths about why enterprises struggle with AI implementations and what separates the toys from the tools. The conversation kicks off with AnswerRocket CEO Alon Goren sharing his journey from pre-PC era computing to building AI solutions that actually work.

The gang dives deep into the current state of AI agents in the workforce. Are they the new interns who need constant training, or are they about to trigger the largest labor disruption of our lifetime? From Klarna's famous (and failed) attempt to replace support staff to the reality of AI's impact on software development, the crew debates whether this time really is different.

Key topics covered:

• Why business problems, not technology, should drive AI adoption

• The real reason companies are freezing hiring (hint: it's not what you think)

• Meta's NBA-level salaries for AI talent and what it means for everyone else

• The "I wish" framework for identifying viable AI use cases • Why the best AI implementations start narrow and expand gradually

Chapters

00:00 Introduction: Meet Alon Goren

06:19 The AI Pilot Dilemma

11:57 Navigating AI Use Cases

21:08 AI Agents: The Largest Disruption of Our Lifetime? 32:05 How to Build Your AI Career

35:43 The AI Talent War: Recruitment and Retention Strategies

41:40 Assessing AI Depth of Expertise in Interviews

Is your enterprise AI pilot part of the 95% that's failing? MIT's latest research just confirmed what many suspected: almost all enterprise AI initiatives are floundering. In this episode, we dig into why companies are hemorrhaging money on AI that never delivers real value, and what the successful 5% are doing differently.

Forget vendor promises and get ready for some uncomfortable truths about why your text-to-SQL dreams might be nightmares waiting to happen.

In this episode, we cover:

• The MIT Bombshell: Only 5% of companies achieve real revenue acceleration from AI pilots. We unpack why the successful few focused on efficiency and cost reduction first (not revenue lift), why the failure isn't about model quality but a massive "learning gap," and what this means for enterprises betting big on AI transformation.
• Why LLMs aren’t like Traditional IT: Working with LLMs isn't like building traditional software—it's like shaping jello. We explain why accepting "like" instead of "equals" is fundamental to AI success, and why the stochastic nature of these systems breaks everything IT departments think they know.
• The AI Text-to-SQL Fantasy: We reveal why text-to-SQL is creating massive business risk, especially when vendors are actively encouraging practices that put companies in danger. Plus, Mike's Czech language disaster that perfectly illustrates why business users + auto-generated SQL = catastrophe.
• From Pilots to Production: Stop thinking "AI project" and start thinking "smart new employee." Jim's framework for AI as onboarding rather than implementation flips the script on why personal productivity with ChatGPT is easy, but enterprise scale is hard. Learn the one question that stops hallucinations and why the successful 5% focus on efficiency, not revenue.
• Real Agents vs. Buzzwords: What actually separates an agent from just another LLM call? Mike's three-point definition cuts through the hype, plus we showcase agents that are delivering real value today (hint: it's not what most vendors are selling).

Follow the Gang:

• Pete Reilly, AnswerRocket, COO - LinkedIn
• Mike Finley, AnswerRocket, CTO - LinkedIn
• Stew Chisam, StellarIQ, Operating Partner - LinkedIn
• Jim Johnson, AnswerRocket, Managing Partner - LinkedIn

Chapters:

00:00 MIT Study: 95% of AI Pilots are Failing

02:07 The 5% That Succeed: Cost Reduction vs. Revenue Lift

03:03 How Internal Bureaucracy Killed a Working AI Pilot

03:55 The Jello Problem: Why LLMs Don't Fit Traditional IT

07:40 Personal Productivity vs Enterprise Scale

11:23 The Complexity of AI Integration

14:05 Treat AI Like A New Employee

16:16 The Stochastic Nature of AI Models

19:48 Risks of AI in SQL Generation

27:22 Making AI Deterministic

29:42 Understanding AI Hallucinations

31:11 What is an Agent, Really?

33:48 The Spectrum of Agent Complexity

38:42 Agents in the Wild: Suno, Lovable, and Deep Research

42:27 Computer Use and the Future of RPA

46:47 MCP Servers and Tools Use

Keywords: enterprise AI failure, MIT study, AI pilots, LLM implementation, AI agents, stochastic models, SQL generation, computer use models, AI hallucination, enterprise transformation

Is your company's AI strategy stuck in the sandbox?  You're not alone. Despite the endless hype, many large companies are finding their AI projects are stuck in the experimental stage. In this episode, we get real about why organizations are struggling and what you can actually do about it.

Forget the hype and join us for a candid discussion on the real-world challenges and opportunities of enterprise AI. We cut through the noise to give you a practical playbook for moving forward.

In this episode, we tackle:

• The "Vibe Coding" Trend: A new term coined by OpenAI's Andrej Karpathy is taking the developer world by storm. We break down what "vibe coding" is and discuss its double-edged sword for the enterprise: Is it a game-changer for rapid prototyping or a source of security risks and "AI slop"?
• The "AI Prevention Department": We explore the common roadblocks stalling AI adoption, from IT departments that lock everything down out of fear to the lack of a proactive, enterprise-wide strategy. Learn why the highest-risk approach is trying to shut it down instead of creating a safe and effective way to enable your teams.
• Getting Unstuck: Stop thinking in terms of "AI projects" and start focusing on business problems and opportunities. Our experts share actionable advice on how to move from experimentation to real value, including focusing on specific use cases and building a flywheel that funds innovation.
• GPT-5, The Real Story: Was the launch of GPT-5 a revolution or just an evolution? Hear our unfiltered first impressions. We discuss why it's "flat smarter" and more contextually aware, but also how its disruptive rollout broke critical workflows—a cautionary tale for any business leaning into AI agents.

Follow the Gang

Pete Reilly, AnswerRocket, COO
Mike Finley, AnswerRocket, CTO

Stew Chisam, StellarIQ, Operating Partner

Jim Johnson, AnswerRocket, Managing Partner

Chapters

00:00 Introduction to Vibe Coding

11:57 The Future of Coding with AI

20:36 Where Enterprises Struggle with AI

23:27 Navigating AI Security Concerns

26:07 Demonstrating The Value of AI

33:07 Getting AI Initiatives Back on Track

40:32 First Impressions on GPT-5

46:17 Comparing User Experiences with AI Models

Keywords: Vibe coding, AI in enterprise, GPT-5, coding tools, AI challenges, productivity, software development, technology trends, coding practices, enterprise solutions