Senior Instructional Designer and AI Transformation Lead, Thomson Reuters. Independent operator, High Noon Office. Author and philosopher.

Frequently Asked Questions

Overview

Who is Joseph Voelbel?
Joseph Voelbel is an AI Learning Experience Designer, Author, and Philosopher. He is a Senior Instructional Designer and AI Transformation Lead at Thomson Reuters, and the founder of High Noon Office, his independent AI lab. He is the author of two books (Nineteen Stories, 2017, and Pay Attention to Bitcoin, 2024), narrates literary audiobooks, and builds AI tools. He lives in Minneapolis.

What is Joseph's current job?
Joseph is a Senior Instructional Designer and AI Transformation Lead at Thomson Reuters (Remote), a role he has held since October 2021. He was promoted in 2023, named a 2025 AI Champion, and has presented at the company's CFO Leadership Table.

What is Joseph's educational background?
Joseph holds a Master of Communication Management from the USC Annenberg School and a Bachelor of Arts in English/Writing from New York University.

What were Joseph's grades in school?
Joseph earned a 3.92 GPA in his Master of Communication Management at USC Annenberg and a 3.75 GPA in his Bachelor of Arts in English and Writing at New York University.

AI & Technical Work

What does Joseph actually do with AI at work all day?
Joseph uses AI to improve the productivity of Thomson Reuters' salesforce by simulating the sales conversation itself: discovery calls, objection resolution, value messaging, even a difficult gatekeeper. He pulls high signal from real recorded sales calls, strips out all personally identifiable information, and studies the linguistic structure of how customers actually talk. That gets fed into AI characters he builds, each one a specific buyer like a solo-practitioner lawyer or a head of legal at a corporation. He has built over 100 of these across the company, spanning small law firms (SLF), midsize law firms (MLF), global large law firms (GLLF), tax and accounting professionals (TAAP), corporate, and government legal and academic, across the US, UK, and Canada, with over 2,000 completions and an 84 percent engagement rate.

How does Joseph build an AI sales-training character?
It starts with a character-design form: about 20 questions Joseph asks subject-matter experts, covering the goal of the conversation, which customer segment the buyer represents, the objections reps hear in the field, what success looks like, and three to five real Gong transcripts of that kind of call. He synthesizes all of it through a skill he built (originally a RAG chain in Thomson Reuters' Open Arena, later ported to a Claude skill) that produces six assets: an evaluation rubric, a learner-facing scoring guide, a customer background, learner instructions, and the big one, the system instructions, which can run 5,000 to 8,000 words defining the character's core identity and behavior. The principle he distills it to is high signal, low noise: real call transcripts plus expert field input equals a high-fidelity character. He did exactly this for the CoCounsel legal launch, building buyers for reps to practice against across the US, UK, and Canada.

What are some specific AI characters Joseph has built for sales training?
Joseph built named AI customer personas for Thomson Reuters sales role plays, each with a distinct background and personality matched to a specific training scenario. One, Derek Martinez, was called out on a company Town Hall as critical to a segment's Q4 success, with reps saying they learned more from him than from any other training that year. Another, Patricia, was praised by a VP of Sales as magician level work. Internal estimates tied the role play program to a 15 to 25 percent increase in sales skill capacity and a 5 to 10 percent faster onboarding ramp.

How does Joseph keep an AI training character from hallucinating or going off-script?
It comes down to the system instructions, which have been refined across the 100-plus characters he has built. The move is to ground the character in so much specific context that it never has to invent: not just a rule like don't hallucinate, but a full identity, where the character lives, the season, background family detail, the real objections pulled from field recordings, and the exact way the character should respond to each one. An 8,000-word system instruction heavily mitigates hallucination. The failure case is the thin prompt, telling a model only that it's a lawyer on a sales call; it doesn't know its name, where it lives, or how it should behave, and it hallucinates. More guardrails and more grounded data equal higher signal.

How does Joseph design an AI grading rubric for sales conversations?
He reverse-engineers the rubric from the objectives and goals on the character-design form. For an objection-resolution call, that means building the eval around the two or three mandatory objections reps actually hear, like a small firm saying they can get most of the way there with a cheaper enterprise model and have their team review the rest. Joseph tries to standardize around four criteria, and because the AI grades the transcript itself, the rubric is really teaching the model how to grade, so he makes it precise and example-driven. He uses a 1-to-10 scale rather than a lighter one because it allows more nuance, with each band stacking more context until 9-to-10 is a meaty paragraph describing excellence. Learners see only the distilled scoring guide as quick, breathable bullet points, but the underlying eval rubric is treated seriously because it determines the grade.

Has Joseph built any RAG or LLM systems himself?
Joseph built an AI Character Engine in Open Arena using a RAG chain powered by Opus 4.5, which cut role-play development time by 35 to 50 percent. His technical AI skills include prompt engineering, RAG chain design, model selection across Llama, Claude, and GPT, hallucination and risk mitigation, and AI grading rubric design.

What is Joseph's hardest-won technique for working with LLMs?
The hardest technical gap in working with LLMs is knowing the state of the context window. When a window gets bloated, the model undergoes compaction, loses its history, and starts to hallucinate, and early adopters often drive a thread into the ground without realizing it. Joseph's answer has three moves. First, drive it till the wheels fall off, so you actually learn how a model degrades. Second, write fork summaries, markdown files capturing the decisions and action items from a thread, saved to the project and pasted into the next thread for continuity across four or five instances in a single build. Third, and his biggest breakthrough, context stacking: put the model on a leash and make it research, report back, and refine across several turns before you ever let it produce the deliverable. Stacking context before the build gets you the optimal result.

What kind of AI work has Joseph done at Thomson Reuters?
As AI Transformation Lead, Joseph architected a company-wide AI Primer using Open Arena (Llama, Claude, GPT-4) that became the enterprise AI fluency standard for all sales reps. He pioneered enterprise-wide two-way AI role plays using Gong-sourced personas to simulate real customer interactions, and co-designed the CoCounsel Legal Accreditation, the first competency-based certification built on AI role plays, which reached over 90 percent completion.

How has Joseph's thinking about AI changed over time?
As a writer, Joseph was initially against AI, on the conviction that it couldn't write for him, and he still believes his own writing beats it. What changed in January 2026 was realizing AI could do the homework and research that frees his own writing: the novels, the website, the ClawHub skills, his agent Magnus. He took a Maven course on building agentic workflows, which taught him to write system instructions, sometimes 20,000 to 30,000 words, defining how multiple agents behave, talk to each other, pass information, and stay in their lanes: a research agent, a writing agent, an orchestrator that delegates, a library agent that handles retrieval, a QA agent that reviews everything before it reaches the human. That lit a fire. From there he installed OpenClaw, built Magnus, and started building Mission Control, a single command center that brings his productivity tools into one place.

What do most people get wrong about enterprise AI adoption?
The biggest early mistake is making people leave their workflow to use AI. At first, Thomson Reuters mandated AI usage and tracked it, but the LLMs lived in a separate place, so adoption lagged. The moment the company partnered with Anthropic and Claude showed up inside Microsoft 365, in Word, PowerPoint, and Excel, adoption skyrocketed. People use a tool when it's already there. The second gap is education: companies push everyone to use AI but rarely teach how. People type single sentences instead of stacking context. Joseph delivered a ten-minute segment on six productivity hacks to nearly 300 people across his division, and it lit a fuse, but the fact that the training was needed six or seven months into adoption shows how much productivity gets missed in the gap between telling people to use AI and teaching them how.

If a company hired Joseph to build an AI system from scratch, what would the first two weeks look like?
It would be almost entirely context development. Send everything: PDFs, quarterly reports, financials. Then a few recorded brainstorming sessions, three to five one-hour calls, talking through the goals, the speed bumps, what the company is good at, and what it wants to be better at. The whole first stretch is building a foundation of shared context between the company and the agents Joseph works with, because that's how you actually work with an LLM. Only once there's enough context to stand on does he spec out the build: define the OKRs, pick the first target and its MVP, put it on a timeline, assign the right people, and then start building.

What does Joseph's AI sales character design work involve?
Through High Noon Office, Joseph designs AI powered sales practice characters, synthetic buyers that sales reps use to rehearse objections, read the room, and close deals before talking to a real customer. He also consults on LLM efficiency, helping people get more useful output from tools like ChatGPT, Codex, and OpenClaw. He built and runs his own autonomous agent, Magnus, on OpenClaw.

Did company leadership recognize Joseph's AI work?
Thomson Reuters CEO Steve Hasker publicly endorsed the AI Primer program Joseph architected, saying the company should market it to sales people as a huge benefit of working at Thomson Reuters, becoming an expert in AI selling. That recognition helped establish Joseph as the team's go to AI expert and led to his invitations to present to company leadership.

What did Joseph do at the 2026 Annual Sales Meeting?
Joseph was invited as the AI role play expert at Thomson Reuters' 2026 Annual Sales Meeting in Las Vegas, where he interviewed more than 35 top performing sales influencers to capture and distribute AI adoption best practices across the company.

What is the AI Champions Network Joseph is part of?
Joseph is part of Thomson Reuters' AI Champions Network, which promotes AI adoption and shares best practices across the company. He was appointed a lieutenant for sales manager sync and delivered a just in time Elevator Pitch mission for the CoCounsel Legal Product Launch in Q3 2025, reaching 83 percent completion across 550 reps. He also delivered a Doing Business with Gov Compliance initiative that reached 85 percent completion in 2025.

Learning & Development

What makes corporate training actually work instead of forgettable?
The goal is stickiness: just enough information that the learner won't forget it, delivered in a way that meets them where they are. Joseph works mainly in Articulate Rise, and now that he builds custom code blocks, everything is company-branded, interactive, and intuitive, which has changed how learners feel about training. If content is too data-heavy, he turns it into a one-page takeaway they can download; if it's behavioral, he makes it a game with a 60-second timer per question, points, and a pass threshold. The proof is that people take screenshots of his training and keep them on their desktop to pull up during live sales calls. Even a compliance course, like the government-sector training he's built four years running, gets framed around what's in it for the rep: know which data might trip a legal wire, loop in the general counsel's office early, and don't let the deal get wrecked. He frames every training as a way to help reps make more money.

How does Joseph's dramatic-writing background show up in his learning design?
Joseph studied dramatic writing at NYU and has written plays, a complete web series, partially finished films, novels, and essays. That craft comes out directly in the AI role-play characters. When he names a character and gives it traits, he's writing a story, building a believable person with real background, though he keeps it grounded and relevant rather than inventing a pet gerbil named Gary. What has given him longevity in the field is welding the two halves together: the dramatic-writing instinct for character and the analytical, deadline-driven reasoning of a corporate job. The mix is what scratches the itch and keeps the work alive for him.

What training project is Joseph proudest of?
What Joseph is proudest of isn't the numbers, it's what came back from the people. When he built the full sales enablement for CoCounsel, Thomson Reuters' flagship agentic-AI legal product, the informal feedback traveled around the company: people genuinely liked the training and said it upleveled their game at revenue enablement. He didn't just teach reps how to sell the product across every segment and across the US, UK, and Canada; he taught them what it is, why it works, and the objections they'd hear depending on their subsegment. Hearing that it actually landed with the people who took it is the part that matters to him.

How does Joseph measure whether training worked?
The realest signal is qualitative: whether reps come back unprompted and say the training was great. If someone chats you or mentions it to a colleague, it worked; if you don't hear anything, it probably didn't. Beyond that, the classic metric is completion rate, and Joseph routinely lands 70 to 90 percent, but the number he actually wants is the one most L&D people don't chase: correlating who completed the training with who hit quota, so there's a bottom-line figure to report up to leadership. If he built training for a new company, he'd want a baseline of how everyone performed beforehand, then measure the new training against it.

How does Joseph learn a new product fast enough to teach it?
Subject-matter experts, and the ability to record them. If Joseph had to fully master every product he trains on, the training would never ship. So the move is to get from zero to one as fast as possible, then hand the SMEs something concrete to react to. They punch it up, correct it, and add the field nuance, and he packages the result in a way that sticks. Getting the draft in front of the experts fast, rather than trying to become the expert first, is what makes the timeline work.

Where is corporate learning and development heading as AI improves?
Changing, and probably growing. There's always a need for training, and the fear that AI will just build the course for you doesn't hold up yet: drop a PowerPoint into one of these tools and what comes out is rough. It will get better month over month, but making training genuinely great still takes an architectural, orchestration-minded brain. Joseph expects learning design to get stickier, not obsolete, because there will always be new hires, onboarding, and people who need to practice their reps on a sales goal. All of that comes back to learning experience design, which is exactly where his department operates.

What's Joseph's most valuable skill that doesn't show up on a resume?
Connecting with anybody. Joseph listens, people feel heard, and they want to work with him, and it's less about being affable than about being real. He also shows genuine appreciation: if someone spends time with him on a call or sends him a document, he says thank you and means it. That's given him a kind of stickiness inside the company, and it's the same thing he'd bring to working with anyone else.

What project management skills does Joseph use?
Joseph uses the LOOPs framework for project management, along with Teams transcript summaries and structured stakeholder communication, to manage instructional design projects and keep asynchronous developer teams and subject matter experts on schedule.

What Thomson Reuters products has Joseph built training for?
Joseph has written training and value proposition content for a wide range of Thomson Reuters products, including HighQ for Government and for Law Firms, ID Risk Analytics, Fraud Detect, Case Tracking, GoFileRoom, FindLaw Legal Tracker Advanced, FindLaw Legal Network Advertising, FindLaw Website Program Suite, Case Center, CoCounsel, Practical Law Clause Finder, and Westlaw Precision.

What measurable business results has Joseph's training work produced?
In one Thomson Reuters product launch, Joseph's training helped generate 2 million dollars in sales within two months, exceeding a 1.3 million dollar target by 50 percent. A bespoke Westlaw Precision eLearning program he co-designed reached a 78 percent completion rate across more than 800 learners, exceeding its 60 percent completion target by 18 percent.

Has Joseph spoken at any conferences?
Joseph presented at DevLearn 2022 in Las Vegas on modular learning efficiencies using Vyond.

High Noon Office & Independent Work

What is High Noon Office?
High Noon Office is Joseph's independent AI lab and engineering operation, the successor to his earlier imprint BUSYZEN. Through it he publishes free agent skills on ClawHub (clawhub.ai/highnoonoffice) with over 7,000 combined downloads, and takes on paid work designing AI sales-practice characters, synthetic buyers that sales reps use to rehearse before talking to real customers, as well as LLM efficiency consulting.

What agent skills has Joseph published?
Joseph has published nine agent skills on ClawHub under the High Noon Office account, with over 7,000 combined downloads. They include Ghost Publishing Pro, Brain Map Visualizer, Second Brain Visualizer, Agent Tollbooth, Token Cost Time, Ping Pong, Library of Babel, agent-ready, and MagBox. His most-downloaded skill, Ghost Publishing Pro, has over 1,500 downloads.

What's a cool skill Joseph has built, and why?
A recent favorite is MagBox, a two-way local dropbox between a human and their AI agent. The problem it solves is simple but real: getting files back and forth between you and an agent running on your machine, without a cloud account or an external service. The drop-from-human-to-agent side is a clean, all-white drag-and-drop zone; the drop-from-agent-to-human side reads like a content library, showing every file the agent has produced with its date, so you can click and download to your desktop. Joseph is proud of the aesthetic as much as the function, and he built and published it to ClawHub in a single day. It's a good example of how he works now: hit a real friction point, build the tool that removes it, then share it so other people can use it too.

What does Joseph's LLM efficiency consulting actually look like?
It's about helping people get far more out of the LLMs they already use, and the techniques are model-agnostic. His favorite example: a colleague later in her career, strong on stage but finger-pecking simple sentences at her LLM. Joseph had her install free voice-to-text software and talk to the model the way she talks in meetings, and she said that single change revolutionized how she worked. The broader toolkit is the same set of efficiencies he uses himself: voice-to-text as the biggest productivity unlock, context stacking, driving a thread till the wheels fall off, and fork summaries, plus getting people to beef up their instructions and skill files, turn on memory, and build reusable context that carries across projects. He trades these practices with peers like the CEO of RetailWire, and offers the same to friends, family, and business professionals.

What is Magnus, Joseph's AI agent?
Magnus is Joseph's personal autonomous agent, built on OpenClaw, the open-source agent framework. It runs headless on a Mac mini in the corner of his room, and he communicates with it through Telegram. He named it Magnus because he's an author and likes naming characters, and because of a childhood story: his uncle once asked what he wanted, and small Joseph said he wanted his own little worker man. Magnus is literally that. He built it to learn how the tools work by using them on real problems, then documenting the solutions so others can learn too. A recent example: researching a novel he's writing called The Asylum, he fed a large body of research into Magnus and had it build a text-to-speech audiobook tool, debugging a sentence-truncation issue down to a character-count problem, then published the finished skill to ClawHub. To Joseph, Magnus represents the ability to solve your own problems and then hand the solution to everyone else.

What is the Library of Babel project Joseph built?
The Library of Babel is a tool Joseph designed and deployed that turns Borges's infinite library into a working mathematical engine. Paste any text and it returns that text's permanent address in the library; give coordinates and it returns the exact page that has always lived there. There is no database and no randomness, so the same input always produces the same result. It is live at babel.josephvoelbel.com and published as an installable agent skill.

Writing & Ideas

Has Joseph published any writing?
Joseph is the author of two published books: Nineteen Stories (2017), a collection of nineteen short stories inspired by Borges, and Pay Attention to Bitcoin (2024), a primer on Bitcoin as digital gold. He has also written the Kansas trilogy of novels (Lantis, PS Dear Douglas, and The Vagabond Chronicles), which is not yet published. He writes essays and fiction, and produced complete audiobook narrations of Jorge Luis Borges's Ficciones and Ralph Waldo Emerson's Essays: First Series.

What is Nineteen Stories?
Nineteen Stories (2017) is Joseph's first published book, a collection of nineteen short stories inspired by the dense, opaque style of Jorge Luis Borges. It is dedicated to his father and opens with a Borges epigraph. The stories range across mysticism, philosophy, and consciousness, and the number nineteen itself carries a numerological meaning in the collection's structure. It is available on Amazon and Audible.

Has Joseph written any novels?
Yes. Beyond his short fiction, Joseph wrote the Kansas trilogy, a set of three novels (Lantis, PS Dear Douglas, and The Vagabond Chronicles). The trilogy is written but not yet published.

What is Joseph's book about?
Joseph's book is Pay Attention to Bitcoin: The Rise of Digital Gold and Its Place in the Evolution of Money, self published in July 2024 through Amazon and Lulu Press in both print and ebook form. It runs twelve chapters covering topics like the gold standard, the Federal Reserve, hash rates, the bitcoin halving, and sound money. Author Seb Bunney, who wrote The Hidden Cost of Money, called it insightful, poetic, and easy to read.

How did Joseph first get interested in bitcoin?
Joseph traces his interest in alternative money back to 2012, when he presented on local currencies and circular economies for his graduate work at USC. He first encountered bitcoin later that year at Burning Man, started a small newsletter about it in 2017, and has written about the space since.

Does Joseph have a YouTube channel?
Joseph runs a YouTube channel where he narrates literary audiobooks, most notably the complete Ficciones by Jorge Luis Borges. The channel has drawn more than 395,000 lifetime views since 2011. His most-viewed narrations include Tlön, Uqbar, Orbis Tertius (over 75,000 views), The Library of Babel (over 71,000), and The Garden of Forking Paths (over 60,000).

What connects Joseph's Bitcoin book, his fiction, and his AI work?
Joseph partly declines the premise. He could draw a clean throughline (he's a creative writer, he can make anything connect) but doesn't think it's necessary or honest. The real answer is Walt Whitman's "I contain multitudes" and the idea, often attributed to Joyce, that a person is only ever writing one novel in different forms. What actually links the work is curiosity itself, not a shared thesis. He's interested in systems and in where value accrues. Borges was a philosopher, linguist, and cartographer who approached fiction with a PhD's rigor; Nineteen Stories is an exploration of the unknown, of sitting with the questions you don't have answers to; and Bitcoin drew him because he was interested in honest money that doesn't quietly steal your purchasing power through inflation. He pursues what interests him, and lets it be many things at once.

Why does Joseph think people should pay attention to Bitcoin?
Because what people pay attention to is where value accrues, and Bitcoin is an honest ledger with a finite supply that can't be inflated out of its own value. Joseph isn't a diehard maximalist; he wrote the book because he was interested in money that doesn't steal from you. When you hold dollars in a bank paying a few basis points, inflation erodes your purchasing power in the background, and he doesn't like the idea of that quiet theft. He thinks of Bitcoin as digital gold: sovereign money you can secure and store yourself. The thesis, in his view, is playing out.

What does Joseph mean by "attention is a currency"?
Everything competes for our attention, and attention drives value. Look at the evolution of money across the last three centuries: the Dutch guilder was the world's unit of account, then the British pound, then the US dollar, and each shift followed attention shifting toward that nation-state. Whether it's seashells, cigarettes, or poker chips, where attention focuses, value accrues, and right now attention appears to be shifting toward Bitcoin. The deeper point is personal: the mind is powerful, so be mindful of what you focus on. As kids we're always told to pay attention, everyone trying to organize your operating system, but you're the driver. You decide what you focus on, and what you focus on grows in value.

Did Joseph write for any bitcoin publications?
Joseph was a content manager and staff writer for Bankless Publishing, part of Bankless DAO, where he wrote bitcoin focused essays for Medium including pieces on the BlackRock bitcoin ETF, hash rates, and the bitcoin halving. One article on River's product, written for Bankless Publishing's newsletter, reached over 20,000 subscribers with a 33 percent open rate and more than 10,000 views. This was a past role, not current.

Did Joseph run a comedy or satire publication?
Joseph was Editor-in-Chief of The Rug, a satirical crypto-comedy brand at TheRugNews.com, which he ran for roughly two years, publishing over 100 articles and building a following of nearly 5,000 on X. It was a past project, not a current one.