AI, Actually

Your AI agents work great for you. Getting your team to use the same AI agents is a completely different problem, and it's one most companies trying to scale AI haven't solved yet. This episode breaks down why single-player AI tools like Claude Code and Codex don't automatically scale into team-wide workflows, and why the fix isn't a smarter model.

What's actually missing is better infrastructure for humans and AI agents to collaborate, and that's where AI harnesses come in.

This episode of AI, Actually looks at what an AI harness actually is, the coordination layer that turns a raw model into a reliable business tool, and when it's worth building a custom one instead of using an out-of-the-box harness from a foundation model provider. The team also covers the vendor lock-in risk built into that choice, gives a candid read on the "pause AI" debate sparked by Dario Amodei's blog post, and reacts to JEV, a new fast AI classifier model already replacing multi-step LLM chains in production.

You'll learn:

• Why scaling AI agents across a team is an infrastructure problem, not a smarter-model problem
• What an AI harness is, and why you're always using one whether you realize it or not
• How to decide when to build a custom AI harness versus using a foundation model provider's default
• The real business risk in the AI regulation debate: vendor concentration, not runaway AI
• A first look at JEV, the fast, cheap AI classifier already replacing multi-step LLM workflows

This episode is designed for: business and technology leaders scaling AI agents across their teams, anyone evaluating build vs. buy for AI infrastructure, and leaders trying to separate real AI regulation risk from the headlines.

https://answerrocket.com/

Follow the Gang

• Stew Chisam, Operating Partner, StellarIQ - https://www.linkedin.com/in/stewart-chisam-7242543/ 
• Mike Finley, Co-Founder, AnswerRocket & StellarIQ  - https://www.linkedin.com/in/mikefinley/ 
• Pete Reilly, Co-Founder, AnswerRocket & StellarIQ - https://www.linkedin.com/in/petereilly 
• Alon Goren, AnswerRocket, CEO - https://www.linkedin.com/in/alon-goren-87889681/ 

Chapters:

00:00     Introduction to AI evolution and key topics
02:00     Single-player vs. multi-player AI: limitations & collaboration
04:37     Workspace infrastructure for AI collaboration
06:30     Challenges in sharing AI agents and scaling
15:41     What AI harnesses do
18:20     Why harnesses make AI reliable
25:03     Building an AI strategy
26:46     Should businesses pause AI adoption?
33:25     Fast AI classification
40:25     How businesses should move forward

#AIAgents #EnterpriseAI #AgenticAI #AIStrategy #AIHarness #AIRegulation #ArtificialIntelligence

Most AI is still billed like a utility. You pay for usage and hope it adds up to something worth the spend.

That model breaks down once an agent starts finishing entire jobs instead of just responding to prompts. The real story: pricing has to be tied to what the agent actually delivers, not how much of it ran.

On this episode of AI, Actually, Jim Johnson, Andy Sweet, Nicole Kosky, and Ada Gil break down why AI agent pricing needs to shift toward value, what that actually requires you to define first, and why most companies aren't set up to measure it yet.

You'll learn:

• Why usage-based pricing breaks down once agents start delivering finished work instead of just activity
• How to define an "atomic unit of output" so you know what you're actually paying for
• Why acceptance criteria have to be defined before you build, not after
• Why you can't let an agent evaluate its own output, and why unchecked human evaluation is risky too
• How adoption, not just accuracy, determines whether you're getting the value you're paying for
• Why model cost should flex by task, not default to the most expensive option every time
  

This episode is designed for enterprise leaders evaluating AI agent vendors, building pricing models for AI initiatives, or making the internal case for how AI spend should be measured and justified.

Learn more about AnswerRocket's enterprise AI solutions: https://answerrocket.com/ 

Speakers:

• Andy Sweet, VP Enterprise AI Solutions, AnswerRocket - https://www.linkedin.com/in/andrewdsweet/ 
• Jim Johnson, President, AnswerRocket - https://www.linkedin.com/in/jim-johnson-bb82451/
• Ada Gil, Director, AI Business Transformation - Practice Lead - https://www.linkedin.com/in/ada-gil-84837648/ 
• Nicole Kosky, Senior Director of Services, AnswerRocket | https://www.linkedin.com/in/nicole-kosky-5b9a3b6/
  

YouTube Chapters:

00:00     Introduction to AI's ROI and Business Impact

05:13     From AI Tools to Agentic Workflows

08:29     Treating Agents Like Interns

10:44     Building Trust and Measuring Good Enough

13:31     Adoption, Pricing, and Business Value

20:29     Scaling AI Value Through Better Workflow Design

#AIAgentPricing #ValueBasedPricing #EnterpriseAI #AIActually #AgenticAI #AIROI #AIStrategy

Managing AI costs has become increasingly difficult. Uber’s CTO made headlines when he shared the company had burned through its entire 2026 AI budget in just 4 months. That's the problem this episode tackles head on. Token prices are falling for a given level of intelligence, but most companies' actual AI spend keeps climbing, because the bar for "good enough" intelligence keeps moving too. Nobody has a clean way to forecast it yet.

In this episode of AI, Actually, Jim Johnson sits down with Shanti Greene, Stew Chisam, and first-time guest Jake Barger to break down what's actually driving enterprise AI costs and what to do about it. The conversation moves from the economics of model pricing into a real case study: a 33x cost reduction on a production agent, achieved in about two weeks.

You'll get insight into:

• Why token prices dropping doesn't mean your AI bill is dropping
• How model routing, caching, and context management quietly drive cost
• The step-by-step process behind a 33x cost reduction on a live enterprise agent
• When open weight and self-hosted models actually make sense, and when they don't
• Practical takeaways for building a repeatable cost optimization process

Follow the Gang:

• Jim Johnson, Managing Partner, AnswerRocket - https://www.linkedin.com/in/jim-johnson-bb82451/ 
• Shanti Greene, Head of Data Science and AI Innovation, AnswerRocket - linkedin.com/in/shantigreene/ 
• Stew Chisam, Operating Partner, StellarIQ - https://www.linkedin.com/in/stewart-chisam-7242543/ 
• Jake Barger, Director of AI & Machine Learning, AnswerRocket | linkedin.com/in/jacob-barger-30b1ab1b9/ 

Chapters
00:00     Introduction and panel overview
02:44     Token Economics: Why Costs Stay High
07:34     Model Routing, Caching & Task Fit
12:22     Continuous Optimization for AI Agents
16:51     The 33x Cost-Reduction Case Study
23:36     Context Management & Smart Escalation
29:28     Open-Weight and Self-Hosted Models
37:22     Enterprise Reliability and Deployment Tradeoffs
44:26     Closing Takeaways: Managing AI Spend

#TokenManagement #AICostOptimization #LLMTokenSpend #ModelRouting #AgenticAICosts #OpenWeightModels #SelfHostedLLM #KVCacheOptimization #EnterpriseAIROI #AIInfrastructure

A tweet from Peter Steinberger asking whether AI builders should still be "talking loops" or "shifting to graphs" hit three million views and split the AI world into camps. Most business leaders have no idea what either term means, or why the answer changes how much your AI systems cost, how reliable they are, and whether you can hand them to a compliance officer with a straight face.

This week on AI, Actually, Pete Reilly sits down with Mike Finley, Shanti Greene, and Stew Chisam to translate the loops-versus-graphs debate out of engineering-speak and into plain business terms. Using examples pulled straight from AnswerRocket's own AI-automated finance close process, the team breaks down when a simple agent is enough, when you need a full graph of coordinated agents, and why the difference determines whether your AI system is auditable, predictable, and affordable at scale.

Find Out:

• What is a loop? Why is it just another type of graph?
• An analogy from finance or an org chart that makes this click
• Why does splitting agent responsibilities improve reliability and cut confusion?
• How do graphs give you auditability, predictability, and cost control that loops can't?
• Why does more capable AI mean more human jobs, not fewer?
• What are the two things IT and business executives should be doing right now?
  

Follow the Gang:

• Pete Reilly, AnswerRocket & StellarIQ, Co-Founder, linkedin.com/in/petereilly
• Mike Finley, AnswerRocket & StellarIQ, Co-Founder, linkedin.com/in/mikefinley/
• Shanti Greene, AnswerRocket, Head of Data Science and AI Innovation, linkedin.com/in/shantigreene/
• Stew Chisam, StellarIQ, Operating Partner, https://www.linkedin.com/in/stewart-chisam-7242543/
  

Chapters:

00:00     Introduction and Overview of AI Workflow Evolution

01:31     Understanding Loops as Cyclic Graphs in AI

03:53     Practical Business Examples of Loops and Graphs

17:47     Orchestration: Coordinating Specialized Agents

20:38     Governance, Auditability, and Predictability

23:01     Speed, Cost, and Process Optimization

26:46     Choosing Between an Agent and a Graph

30:02     How AI Changes Human Work

35:47     Safeguards and Sandboxes for Enterprise AI

#AIAgentArchitecture #AgenticWorkflows #LoopsVsGraphs #AIOrchestration #EnterpriseAIImplementation #MultiAgentSystems #AIGovernance #AICostControl #AgenticAIExplained #AISandboxStrategy

The token bills are arriving, and a lot of teams do not like what they see. Halfway through 2026, the promise of cheap, abundant intelligence is colliding with enterprise reality.

Behind the headlines about new models and rising prices is a simpler story about whether your AI spend is actually buying business value. In this episode of AI, Actually, Jim Johnson, Nicole Kosky, Shanti Greene, and Andy Sweet connect three topics that look unrelated on the surface: early results from Fable 5, the real cost of tokens and total cost of ownership, and how to think about AI ROI.

The throughline is one piece of math. Grow the numerator, scope the denominator honestly, and build the business case before you deploy.

Topics covered:

• Early field results from Fable 5 and why not all tokens are created equal
• Token pricing, the end of the subsidy era, and the surprise-bill problem
• What actually belongs in the denominator of an AI business case
• The human in the loop, AgentOps, and the trust gap that paces adoption

• Shanti Greene, Head of Data Science and AI Innovation, AnswerRocket - https://www.linkedin.com/in/shantigreene/ 
• Andy Sweet, VP Enterprise AI Solutions, AnswerRocket - https://www.linkedin.com/in/andrewdsweet/ 
• Jim Johnson, President, AnswerRocket - https://www.linkedin.com/in/jim-johnson-bb82451/
• Nicole Kosky, Senior Director of Services, AnswerRocket - https://www.linkedin.com/in/nicole-kosky-5b9a3b6/

Learn more about AnswerRocket: answerrocket.com

Chapters:

00:00     Introduction to AI topics and guest insights

02:20     Experimenting with Fable 5 and its value

04:00     Understanding token consumption and TCL

05:34     Cost and ROI considerations in AI models

07:52     The role of model reasoning and creativity

10:07     Managing costs and change in AI deployment

13:28     Strategic planning and business case development

15:02     Evolving models and maximizing value

16:51     Testing, evaluation, and human-in-the-loop

18:53     Leadership and organizational adoption of AI

21:15     The "Trust Gap" for Adoption

25:27     Closing thoughts and industry outlook

#EnterpriseAI #AIROI #AgenticAI #AIStrategy #LLM #AgentOps #AIActually

Everyone is waiting for the next model to make their AI problems disappear. The teams getting real value already figured out the model was never the hard part.

In this episode, the crew reacts to Nate B. Jones and his argument that the trillion dollar opportunity sits in completed workflows, not smarter chat.

This is AI, Actually, the show that cuts through the hype and gets practical about putting AI to work. We dig into why the model alone will never get you there, what happens to SaaS in an agentic world, and how to spot the workflows actually worth automating.

You'll learn:

• Why completed workflows, not models, are where enterprise value actually lands
• How to use evaluation and a clear definition of "what good looks like" as the starting point for any build
• What forward deployed engineers really do, and why the business person beside them matters just as much
• Why seat-based SaaS pricing is breaking and what replaces it
• How to pick a first workflow to automate using repeatability, ROI, and human in the loop
• What harnesses are and why they are becoming the next layer of value
  

Learn more about AnswerRocket: https://answerrocket.com/ 

Get Ethan Mollick’s book, “Co-Intelligence: Living and Working with AI” here: https://a.co/d/0jl6kAMd  

Follow the Gang:

• Shanti Greene, Head of Data Science and AI Innovation, AnswerRocket - https://www.linkedin.com/in/shantigreene/ 
• Mike Finley, Co-Founder, AnswerRocket & StellarIQ  - https://www.linkedin.com/in/mikefinley/ 
• Pete Reilly, Co-Founder, AnswerRocket & StellarIQ - https://www.linkedin.com/in/petereilly 
• Ada Gil, Director, AI Business Transformation - Practice Lead - https://www.linkedin.com/in/ada-gil-84837648/ 

Chapters:
00:00     Introduction to AI Business Transformation
03:54     The Model Alone Isn't Enough
08:21     AI Won't Be Implementing Itself
11:43     Rebuilding Workflows and Embracing Change
13:25     The Future of SaaS and AI Integration
18:24     Navigating the Expenses of SaaS Platforms
23:14     The Evolution of Standards in AI
25:34     Understanding Workflows for Automation
28:34     Defining Success in Automation
32:35     The Value of Forward Deployed Engineers
34:47     Identifying Valuable Workflows
38:41     Economic Considerations in Automation
40:30     Concluding Thoughts & Harnessing AI
43:56     Evaluating Agent Performance
45:55     Combining Domain Expertise with Technology

#EnterpriseAI #AgenticAI #AIStrategy #AIAgents #CompletedWorkflows #ForwardDeployedEngineers #AIImplementation

Most enterprise AI projects don't fail because of the model. They fail because of data, specifically, the gap between having data and having data an AI can actually understand and act on. This is the last mile problem, and it's quietly killing AI ROI across the enterprise.

That's the real story behind the stat: 87% of AI projects never reach production.

This is Episode 20 of AI, Actually, the podcast that cuts through the hype to help business leaders get real value from AI. Jim Johnson is joined by Andy Sweet, Nicole Kosky, and Ben Titmus, who leads the data and infrastructure practice at AnswerRocket, to break down what it actually takes to make enterprise AI work at the data layer.

You'll learn:

• What "data readiness" actually means for AI, and why it's different from what your BI dashboards needed
• How to define and build a semantic layer iteratively, without a massive multi-year initiative
• Why throwing everything at an LLM produces confident-sounding wrong answers, and how guardrails prevent it
• How Microsoft Fabric, Snowflake, Databricks, and BigQuery are each approaching semantic layer infrastructure
• Why model-swappability, not model loyalty, is the right architecture posture right now
• What Chief Data Officers should do in the next 90 days to lead AI transformation from the front

This episode is designed for: Chief Data Officers, VP Analytics, data engineering leaders, and any executive responsible for making AI work inside a real business.

Learn more about AnswerRocket's enterprise AI solutions: https://answerrocket.com/ 

Speakers:

• Andy Sweet, VP Enterprise AI Solutions, AnswerRocket - https://www.linkedin.com/in/andrewdsweet/ 
• Jim Johnson, President, AnswerRocket - https://www.linkedin.com/in/jim-johnson-bb82451/
• Nicole Kosky, Senior Director of Services, AnswerRocket - https://www.linkedin.com/in/nicole-kosky-5b9a3b6/
• Ben Titmus, Senior Director, Practice Leader - AI Data, Platforms and Infrastructure - https://www.linkedin.com/in/benjamin-titmus-a2817626/ 

Chapters:

00:00     Introduction and Milestones

01:40     The Data Dilemma in AI

02:59     Understanding Data Readiness

05:50     Defining the Semantic Layer

08:46     The Importance of Context in AI

13:08     Navigating AI's Limitations

16:44     Building Guardrails for AI

21:12     Achieving ROI in AI Projects

26:45     Recommendations for Chief Data Officers

#EnterpriseAI #AIStrategy #DataReadiness #SemanticLayer #AIActually #ChiefDataOfficer #AIAgents

Popularized by Palantir, the term "Forward Deployed Engineer" is everywhere right now.But what does it actually mean, and is it even the right label? In Episode 19, the AI, Actually crew digs into one of the hottest buzzwords in enterprise tech, separating the signal from the noise and getting real about the skills, structures, and organizational shifts that matter.

This is a candid conversation about how AI is collapsing traditional role boundaries, and why the ability to understand a business problem may now be more valuable than the ability to write code. The crew also tackles why large enterprises are structurally misaligned with how this work actually gets done, and what education, apprenticeship, and org change need to look like going forward.

Topics covered:

• What "Forward Deployed Engineer" really means, and why "Forward Deployed Consultant" might be more accurate
• Why senior engineers are thriving and junior roles are evolving, not disappearing
• How AI is inverting the old software architecture model: solve fast, clean up later
• Organizational bottlenecks: why most companies are still structured around the wrong constraints
• The SAP paradox and why business process knowledge is now worth its weight in gold
• What universities and apprenticeship models need to look like for the AI era
  

Speakers:

• Nicole Kosky, Senior Director of Services, AnswerRocket - https://www.linkedin.com/in/nicole-kosky-5b9a3b6/
• Stew Chisam, Operating Partner, StellarIQ - https://www.linkedin.com/in/stewart-chisam-7242543/ 
• Jim Johnson, Managing Partner, AnswerRocket - https://www.linkedin.com/in/jim-johnson-bb82451/
• Shanti Greene, Head of Data Science and AI Innovation, AnswerRocket - https://www.linkedin.com/in/shantigreene/

Chapters: 

00:00     Introduction

01:43     Understanding the Role of Forward Deployed Engineers

04:16     The Evolution of Software Engineering Roles

07:22     The Business Analyst vs. Forward Deployed Engineer

10:47     Organizational Challenges in Adapting to New Roles

12:10     The Importance of Clarity in Problem Solving

17:43     Navigating Change Management in Organizations

20:44     The Role of AI in Redefining Business Processes

28:42     The Future of Software Engineering and Business Skills

32:25     The Need for Educational Reform in AI Skills

37:58     Conclusion: Embracing the Renaissance Person in Tech

#ForwardDeployedEngineer #EnterpriseAI #AIRoles #BusinessAnalyst #TechnicalConsultant #AIOrgChange #AIWorkforce #AgenticDevelopment #EnterpriseSoftware #AIAdoption

Most companies are announcing AI strategies and running pilots that go nowhere. Meanwhile, individuals using AI every day are quietly getting 10x leverage on their work, and the gap is widening.

The real story isn't about tools or models. It's about people, habits, resistance, and what it actually takes to build a culture where AI creates compounding value.

In this episode of AI, Actually, host Pete Reilly is joined by Mike Finley, Nicole Kosky, and new AnswerRocket team member Michelle Hamilton, Director of AI Adoption and Change Management, for a candid conversation on how to move past fear, shadow AI, unused licenses, and the "is this going to replace me?" question that's quietly stalling enterprise adoption everywhere.

You'll learn:

• Why a people-first approach to AI matters more than which model you pick
• How to use "meta prompting" to get dramatically better outputs across any model
• What a personal AI "corpus" is and why it becomes your most valuable competitive asset
• Why the most successful AI adopters are building tools, not just using tools
• How to think about the crawl/walk/run framework for individuals and organizations
• What forward deployed engineers are and why everyone can be one now

This episode is designed for: Business leaders, team managers, and anyone who knows they should be using AI more but isn't sure where to start.

AnswerRocket: https://answerrocket.com/

Speakers:

• Nicole Kosky, Senior Director of Services, AnswerRocket | https://www.linkedin.com/in/nicole-kosky-5b9a3b6/
• Mike Finley, CTO, AnswerRocket  - https://www.linkedin.com/in/mikefinley/ 
• Pete Reilly, COO, AnswerRocket - https://www.linkedin.com/in/petereilly 
• Michelle Hamilton, Director, AI Adoption and Change Management - https://www.linkedin.com/in/michellehamiltonai/ 

Chapters: 

00:00 Navigating the AI Landscape

01:25     Embracing a People-First Approach to AI

05:45     Tools and Techniques for AI Adoption

08:36     Meta Prompting and Leveraging AI Models

11:38     The Challenge of Change in AI Integration

14:33     Collaboration and Knowledge Sharing in AI

17:06     Empowering Non-Technical Users with AI

20:18     The Future of AI and Human Collaboration

27:58     Using AI for Office Tasks

33:18     The Role of AI in Education

36:42     Evolving Workflows with AI

45:35     Getting Started with AI

#AIAdoption #EnterpriseAI #AIActually #ChangeManagement #GenerativeAI #AIStrategy #FutureOfWork

Every week, someone declares that AI will make consultants obsolete. What we actually see is AI breathing new life into consulting, with practically every model maker teaming up with consulting firms for delivery support.

The real story isn't about whether consulting is dead. It's about whether the consulting model your company is paying for is built for this era or the last one.

In this episode of AI, Actually, Jim Johnson hosts Evan Gatch, Nicole Kosky, and Andy Sweet for a candid conversation on what enterprise AI consulting actually looks like right now: what's working, what's failing, and what clients should be demanding from their partners.

You'll learn:

• Why IT-led AI initiatives consistently underperform and what the Fortune 500 failures have in common
• How the consulting business model is shifting from hourly rates to outcome-based and gain-share agreements
• Why prototypes are easier to build than ever and why that makes the POC-to-production gap more dangerous, not less
• What "enterprise grade" actually means when it comes to agentic AI solutions
• How to structure a use case prioritization framework before you start building
• Why organizational change management is the constraint that technology alone can't solve

This episode is designed for: CDOs, VP-level analytics and data science leaders, AI transformation leads, and enterprise consulting buyers navigating AI strategy decisions.

Learn more about AnswerRocket's enterprise AI solutions: https://answerrocket.com/

Speakers

• Andy Sweet, VP Enterprise AI Solutions, AnswerRocket - https://www.linkedin.com/in/andrewdsweet/ 
• Jim Johnson, President, AnswerRocket | https://www.linkedin.com/in/jim-johnson-bb82451/
• Nicole Kosky, Senior Director of Services, AnswerRocket | https://www.linkedin.com/in/nicole-kosky-5b9a3b6/
• Evan Gatch, VP Consulting Sales, AnswerRocket - https://www.linkedin.com/in/evan-gatch-1531482/ 

Chapters

00:00 Introduction and Meet the Experts

02:10 Consulting in the Rapidly Evolving AI Landscape

04:14 The Changing Nature of Consulting Services

07:18 Aligning AI Solutions with Business Goals

14:02 The Role of Software Engineering Skills in AI

17:17 Organizational Change and AI Adoption Challenges

21:42 Enterprise-Grade AI Solutions and Production Readiness

29:28 Structured Approaches to AI Opportunities

32:45 Outcome-Based Pricing and Trust in Consulting

38:29 Closing Remarks and Future Outlook

#AIConsulting #EnterpriseAI #AIStrategy #OutcomeBased #AgenticAI #AITransformation #AIActually