Claude Code for Absolute Beginners — What We’ve Learned Teaching AI to Non-Coders

Kate Lee, editor-in-chief of Every, sits down with Mike Taylor — Every’s newly full-time Head of Tech Consulting — to discuss what they’ve learned from teaching Claude Code to absolute beginners. Every is a subscription media company offering a newsletter, five productivity apps (as of the week of this stream), and training programs for both enterprise and beginner-level students. Taylor, a former marketing agency founder who taught himself to code over five years of evenings and weekends, brings a practitioner’s perspective on AI adoption: he runs multiple AI agents daily, has taught prompt engineering to over 300,000 students on Udemy, and now designs live training for both beginners and enterprise executives. The conversation reveals that the biggest barriers to AI adoption aren’t technical — they’re emotional, linguistic, and structural — and that context management is the single skill that separates effective AI users from frustrated ones.

The language barrier is real. Taylor describes the experience of non-technical users encountering developer terminology as “almost like landing in a foreign country and everyone’s speaking a language you don’t understand.” Concepts that seem straightforward once understood — like why it’s called a “pull request” when you’re pushing code — are anachronisms that intimidate newcomers. That particular example came up during an actual cohort, a live teaching moment that illustrates just how many traps the developer lexicon sets for beginners. Developers, Taylor argues, get paid well partly because they’re willing to tolerate this complexity. The good news: the tooling is catching up. Teaching has already shifted from terminal-only to the desktop app, and 80–90% of what users need can now be done without ever looking at code.

Meet the Instructor: From Marketing to AI Consulting

Mike Taylor joined Every full-time as Head of Tech Consulting after writing the “Also True for Humans” column for the past couple of years. It had only been a couple of weeks at the time of this stream, but he’d already been to San Francisco and Miami and completed a bunch of consulting projects — plus a few more over December and January before officially starting. (He opens the conversation by cringing on-camera: he noticed he’d capitalized “Tech Consulting” in his title and knew Kate, as editor-in-chief, would hate it. “Warms my heart,” she replies.)

Every’s consulting team is intentionally small:

Natalyia runs consulting overall
Mike Taylor handles the tech side — working with engineering teams at tech companies plus operations and marketing teams on coding agents and knowledge work automation
Brooker Belcort covers the finance vertical — his background includes leading the finance vertical at Perplexity, working with hedge funds and investment banks

Taylor’s own coding journey started pre-AI. He ran a marketing agency, spent five years doing bootcamps and free MIT/Harvard courses on evenings and weekends, then quit to code full-time for about six months before building anything useful. That was 2020 — just in time to be technical enough to make API calls to GPT-3. He’s been freelancing as a prompt engineer for the past five years — though the role of “prompt engineer” has morphed and changed over that period, becoming “a little bit more horizontal than vertical” as a skill set across different jobs and roles rather than a standalone position. He currently runs about three AI agents around the clock for various tasks. He acknowledges being technical gives an advantage with these tools — they’re not always beginner-friendly — but insists you don’t need to know how to code to use Claude Code. You do need someone to take you through it: “This is the terminal, here’s how you navigate to a folder, here’s what Claude is doing under the hood.” Once past that hump, Claude itself becomes the instructor.

The Three Paradigm Shifts in AI

Taylor identifies three major leaps in how people interact with AI, each as significant as the last:

No AI → ChatGPT: The introduction of conversational AI as a general-purpose tool
ChatGPT → Copilot: AI embedded in your documents — it can see what you’re working on and collaborate in context. This includes both GitHub Copilot for code and the broader copilot paradigm across productivity apps (Google Gemini in Docs, etc.)
Copilot → Agentic AI (Claude Code): An AI that operates on your computer autonomously. The difference is fundamental — it’s like managing an employee rather than collaborating with a peer. Codex and the Google Gemini CLI occupy this same tier

The critical insight: these steps shouldn’t be skipped. Someone who hasn’t used ChatGPT regularly shouldn’t jump straight to Claude Code. Someone who hasn’t used copilot-style tools in their documents may find agentic AI too large a leap. Taylor recommends allowing yourself the natural progression, even though it feels slow, because skipping steps leads to overwhelm and abandonment. “It is totally fine to be wherever you are on the diagram. The important thing is that you just know where you are.” He had the benefit of progressing incrementally starting in 2020 — making API calls to GPT-3, then ChatGPT, then copilots — so the transitions made sense. For someone entering now, the temptation to skip is strong, but the risk of spinning out is real. “Just allow yourself that progression as well,” he advises. “Otherwise, I think it’s pretty easy to spin out and feel like you’re overwhelmed and not getting value from the material that you’re looking at.”

Taylor hints at further paradigms beyond agentic AI that are “more experimental” but doesn’t elaborate. He notes that the really advanced material — agent swarms, multi-agent orchestration — “doesn’t exist anywhere actually. It’s in person and people are talking about it in events.” You can’t discuss agent swarms “in the same breath” as introducing someone to the terminal. Taylor is working on getting more of that material into courses and his column, but for now the frontier lives in hallway conversations and meetups, not curricula.

Is Coding Dead?

Taylor’s answer is nuanced. In a literal sense, yes — he doesn’t write code anymore. His day-to-day role is closer to product manager or engineering manager: reading and reviewing code rather than writing it. Many early-adopter companies report AI generating 100% of their code.

However, the purpose of coding — solving problems — will never die. Taylor learned to code because he couldn’t afford a developer. Now he has a developer that costs $200/month, doesn’t sleep, and writes better code than he ever could.

The people struggling most are those who coded for the craft. They’re mourning a loss. Those who coded as a means to an end — solving problems, building things — are thriving because the bottleneck has shifted from implementation to ideation.

Legacy codebases remain an exception. When the codebase is deeply legacy and the AI lacks organizational context about historical decisions, manual coding still makes sense.

What the Beginners Course Actually Looks Like

The course promise: learn Claude Code in one day. Taylor acknowledges this wouldn’t have been possible even a couple years ago — he references a company called Decoded that used to offer “learn to code in a day” courses. They were great, and to their credit, students did put something live on the internet by the end of the day. But what they actually learned was what JavaScript, HTML, and CSS were — not how to code independently. As Taylor puts it: “You could learn how to code in a day, but you could also learn how to code in 10 years. It depends on what you mean by learn how to code.” The difference with Claude Code: by end of day, students genuinely have a working project and — more importantly — the ability to keep building without an instructor.

Structure

Setup phase: Getting Claude Code installed and running, with TAs swarming the chat and pulling people into breakout rooms for individual troubleshooting. Something always goes wrong with 100 students on different machines
~50% build time: Students work on their own projects with guidance
Personalized challenges: The course increasingly offers a wider selection of challenges rather than one-size-fits-all. For executive offsites, Taylor uses Claude to generate three custom project ideas personalized to each participant’s job role — something he’d never have time to do manually
Custom data: If a student wants to build a Salesforce workflow but doesn’t have Salesforce access during the session, the team prepares example Salesforce data that works the same way once real connectors are in place
Prerequisite: Ideally, students should have Claude Code installed and have prompted it at least once (“Hello, what’s the weather in Hoboken?”)
Target audience: Between “literally never opened it” and “spent more than 10 hours” — roughly under one hour of prior experience. Taylor is explicit that if you’ve already built a product or automated a workflow with Claude, you’ve “aged out” of this course

The Course Is Always Changing

The course evolves after every single cohort — partly because Claude keeps releasing new features that reshape what’s possible, and partly because live teaching has been a formative experience for Taylor himself. His previous courses were all recorded and asynchronous — 300,000 Udemy students he never met. Teaching live is “a very different skill set” that gave him “a lot more empathy for going slower” and recognizing things that are “not obvious in any way” to newcomers. What works in one session may be outdated by the next.

Desktop App vs. Terminal

The course has already evolved significantly. A few months ago, everything was taught in the terminal. Now the desktop app handles most of what beginners need. Kate Lee notes that the terminal — native to developers, alien to everyone else — was a major hump, and removing it as a requirement changes the accessibility equation. She counts herself among people who are “now regularly in the terminal” but weren’t not so long ago. Taylor adds that you can do 80–90% of Claude Code work in the desktop app without looking at code.

That said, Taylor has been surprised by the willingness of non-technical people — including C-suite executives at corporate offsites — to get into the terminal once they see the value.

For those wondering whether to use the web version or the desktop app: Taylor recommends the desktop app specifically because it can work in a local folder on your computer. This enables persistence across sessions — you can do a piece of work, save it to the folder, start a new session, and tell Claude to pick up where it left off. Without the desktop app, you’d need to learn GitHub (which even senior engineers mess up) and deal with pull request etiquette — a whole layer of developer workflow that beginners shouldn’t have to navigate.

Common Use Cases Students Want to Build

Organize my chaos: Connect to Gmail, Slack, and calendar — “tell me what I’m doing today,” “prepare me for my meetings, where do I need to go and when,” “what do I need to look at in Slack.” Essentially replacing an EA. Every’s own consulting team uses Claude this way since they don’t have any ops people on the team
Financial data analysis: Custom dashboards from exported data. Taylor notes this used to require a team of two full-time data scientists taking weeks; now he can build dashboards live on client calls. “What’s the NPS score from the latest survey?” — ask Claude, and it’s there
Workflow automation: Salesforce workflows, email processing, and similar operational tasks

The Real Output

By end of day, students have a working project and — more importantly — a mindset shift. Once comfortable with Claude Code, they develop the habit of asking it to do things before jumping to chat or doing tasks manually. Claude then becomes their instructor for everything that follows.

Taylor is candid: all the information is available online for free, just as Harvard’s computer science lectures are free online. The reason to take the course isn’t secret knowledge — it’s the activation energy of being in a room full of people trying the same thing, with instructors who care about your outcomes.

The Gap Between Beginner and Advanced

A viewer asks what sits between the beginner course (under one hour of Claude Code experience) and the advanced course (10+ hours). Taylor’s answer: “Pretty much anything could be taught in five minutes or five years or somewhere in between. It just depends on how deep you want to go.” He’s confident he could fill a five-year university degree curriculum on Claude Code alone, because to really understand it, “you have to understand almost all of software development and maybe even all knowledge work.”

The practical difference in the full-day format is build time. The biggest friction point Taylor hears from consulting clients isn’t capability — it’s time. “They don’t have time to play with this stuff. They don’t have time to save themselves time.” A full-day workshop forces protected time to actually build, which is something most people in their day jobs say they can never find.

Taylor and Kate both observe that you genuinely cannot speak to beginners and advanced users in the same breath. Taylor runs three AI agents 24/7, uses Kieran’s compound engineering plugin to scale development work — that’s where he lives day-to-day. He’d tell all his friends not to join the beginner course specifically because “you cannot talk about agent swarms in the same breath as you’re talking about ‘hey, meet the terminal.’”

Advice for Technical Non-Adopters

A viewer asks about people who have built agents in Copilot Studio, GPTs in ChatGPT, and websites on GitHub — but still haven’t adopted Claude Code specifically. Taylor offers a three-tier recommendation:

Just install it and start playing. If you’re already technical, the fastest path is trial and error. When it fails, look at the code it wrote, figure out what went wrong (usually a missing context or access issue), and adjust
Make friends who’ve gotten over the hump. Taylor says he’s learned enormous amounts from people he’s met at events, in the office, and online. He admits to being “in a really bad habit of meeting strangers from the internet” since moving to New York — following people on Twitter for a while, then reaching out to meet in person
Get one-on-one tutoring. Taylor himself paid a former Flatiron School instructor $60/ h o u r f or ab o u t 10-12 h o u r so f c u s t o m d a t aba se in s t r u c t i o n — r o ug h l y$ 600 total. For someone with specific technical gaps, that kind of targeted tutoring can be more effective than a broad course

He’s explicit: he wants people to take the course only if it’s genuinely right for them.

The Persistence Gap

When asked what separates people who ship projects from those who stall out, Taylor’s answer is unequivocal: persistence, not technical skill.

“There’s no such thing as a technical person, just a persistent one.”

He tried and failed to learn coding multiple times before a three-month full-time stint during COVID forced it. The key attitude shift: learning to be happy when you see an error message, because it means you’re about to learn something. Taylor credits a friend, Rob Desmond, who would respond to every error with genuine excitement — an attitude Taylor initially found infuriating but eventually internalized.

The same applies to Claude Code: when it errors, you send the error back to Claude and ask what it means. You’re in constant dialogue. The persistent loop of try → fail → ask → learn is the entire process.

Security and Trust: Building It Incrementally

Taylor personally runs Claude Code with root access and dangerously skipped permissions — but only after building trust over a long time. His framework mirrors employee onboarding:

Day one: You wouldn’t give a new hire a corporate bank account. Similarly, start with guardrails and developer sandboxes that restrict Claude’s access to a contained part of your system. Taylor draws from his own career: at his first marketing job, it took three months to get “publish privileges” for Google AdWords — he could only send proposed campaigns to his boss for approval
Build trust gradually: As you see consistent good behavior, expand access. Taylor compares it to his first ride in a Waymo self-driving car: “Oh my god, this is kind of terrifying — I’m about to get in a car that’s driving itself.” But within ten minutes he was checking his phone and not even looking at the road. “It felt so safe.” He feels the same way about Claude now. Kate adds that Every just onboarded new colleagues that very day and did extensive preparation — “which is what you would do with a human but is not necessarily something we’ve thought about with an AI” — reinforcing that managing AI agents benefits from the same onboarding discipline applied to humans
Use AI to police AI: For sensitive projects (healthcare clients, production codebases), use more AI to solve the problems AI causes. Taylor’s concrete technique: open a second Claude session and say, “An intern on my team just wrote this code — can you review it?” It might catch things a human reviewer would miss. At scale, Taylor references Kieran’s “compound engineering” plugin that deploys ~12 sub-agents, including a security expert and language-specific best practices reviewer. The result: Taylor trusts this multi-agent reviewed code more than code he reviews solo
Brand police pattern: Some companies run a “brand police” AI that checks every email or ad before publishing for brand compliance

The core insight: don’t view AI as the cause of security problems — it can also be the solution.

Context: The Universal Mistake

Almost every mistake Taylor sees comes down to context:

Too much context: Filling up the session with information causes “context rot” — the AI gets distracted, confused, and produces conflicting outputs. This is analogous to overwhelming a human with too many instructions
Too little context: Not providing what you’d give an intern to do the same work forces the AI to guess, leading to poor results

This applies identically to humans — which is the premise of Taylor’s column name, “Also True for Humans.”

Keeping Business Context Updated

For teams using Claude Code, keeping shared business and strategy context current is the “million-dollar question.” The tricky part: “You only really know that the context is stale when the AI makes a mistake.” And even then, you need the right context to diagnose and fix it.

Taylor’s recommendation: favor connectors over content. Rather than maintaining static documents that inevitably go stale (a problem Taylor experienced at his marketing agency, where SOPs were constantly outdated — team members would skip them because “it doesn’t really work that way anymore” and never update them), connect Claude directly to live data sources via MCP or API integrations. Pull context from the source of truth in real-time rather than maintaining a copy.

As an example, Taylor mentions Every’s internal MCP: rather than downloading and maintaining a repository of all his published articles, he can simply ask the MCP to pull his latest posts directly.

Keeping Up with the Pace of Change

Taylor’s grounding exercise: walk down any street in Brooklyn and ask what’s visibly different from the 1980s. The answer is basically just iPhones. Despite massive technological change with the internet and everything else, very few things have been truly impactful in day-to-day life.

Apply the same filter to AI: look for what people actually keep coming back to, what’s become a core part of their workflow — not what’s trending today. Through that lens, only about ten genuinely interesting things have happened in AI over the past five years. There are many failed projects, many things that seemed impactful at the time but weren’t. Everything else is noise.

At worst, this approach puts you a couple months behind the frontier, which isn’t bad at all.

The Underrated Approach: Try Everything, Note the Failures

Taylor pushes back on the common question “What should I use AI for?” — the framing is wrong. The better question: What can’t you use it for?

His method:

Try to use AI for everything
Note where it fails
When the next model drops, retry the failures
Those failure areas are where humans remain employed

Current example: PowerPoint generation is still poor enough that Taylor would need weeks of optimization and template modification to make it work. But in six months, the problem may simply disappear. He’s genuinely happy when he finds something Claude is bad at — those failures become content for Every.

Creative Professionals: The Surprise Factor

For people coming from creative fields — music, writing, design — the biggest surprise is simply that Claude Code can help them at all. Knowledge workers have closer analogues (Excel macros, working with tech teams on dashboards), but creatives aren’t used to solving problems by talking to an AI. It’s “a little bit more alien” for them.

Taylor sees this as the current “jagged edge” — humans still have better taste than Claude, but Claude knows all the APIs to programmatically interact with music software, design tools, and creative platforms. The combination of human taste with AI execution capability is where the real potential lies, and creatives are typically positively surprised once they see it in action. Taylor actively encourages people with “less corporate” interests to try Claude Code — seeing someone make music with it is more exciting to him than another inbox organizer.

How Executives Adopt Differently

Taylor teaches executives much of the same material as beginners, because he believes they need to see firsthand what their people are dealing with. Now that “everyone is a manager of AIs,” people who are already managers are actually well-positioned — they have the skill set already. But the adoption pattern is distinct:

Slower to start: Executives are rightfully skeptical — and for good reason. If they believed the hype fully, the implications for their entire company would be radical. So they hold back
Faster to execute once convinced: But once they try it themselves and can see through the hype — “Ah okay, I get it. It is actually pretty amazing, but it can’t do this, this, and this yet, and we need to upskill here, here, and here” — they move aggressively and quickly get into action
Natural advantage in AI management: The skills executives already have — context switching, briefing, evaluation, performance metrics — are exactly the skills needed to manage AI agents effectively

Taylor references Dan (likely Dan Shipper, Every’s CEO) and Toby Lütke of Shopify as examples of executives who are “vibe coding.” Lütke is now committing more code personally than he did when he was solo-coding Shopify before becoming CEO of a publicly traded company.

The Future: More AI Coworkers Than Human Ones

Taylor offers his forecast with a caveat: he’s been able to see where things are going since GPT-3 in 2020, but the hard part is timing — whether something takes six months, one year, or two years. And if you follow the current trends to their logical conclusion, “you get into this weird place where you’re talking about utopia or dystopia… the answer is probably somewhere in between.”

With that framing, his forecast: soon, most people will work alongside more AI coworkers than human ones. Importantly, he doesn’t believe this will be driven by mass layoffs. He hasn’t seen anyone legitimately fired because of AI — companies announcing AI-driven layoffs are mostly covering dysfunction. “It’s a good excuse — like firing people because of the economy. There are other people hiring in the same economy.”

Instead, AI coworkers will proliferate as they become more stable and usable — whether through Claude Code, OpenAI, or something else entirely. The shift will be from managing one AI tool to managing entire teams of AI agents — a new form of management science. Learning to manage AI agents is just as difficult as learning to manage humans — and nobody has had that management training yet.

Why Live Courses Beat Recordings

Every intentionally doesn’t sell recordings of their $1, 000 C l a u d e C o d eco u r se . A c u s t o m er ser v i ce in q u i r y c am e in : so m eo n e in E u r o p e w an t e d t o b u y t h eco u r se b u t co u l d n^{'} t a tt e n d l i v e b ec a u seo f t h e t im e d i f f er e n ce . T h ey w er e w i l l in g t o p a y t h e f u l l$ 1,000 — they just wanted the recordings. The team’s response was unanimous and uncoordinated: “I don’t think they should buy the course.” Taylor was proud of the alignment — everyone felt the same way independently, without discussing it first.

Taylor’s reasoning, informed by creating 24+ hours of recorded Udemy content:

Recorded content works when you’re already bought in — your boss told you to learn LangChain, your job depends on it, you’ll pay attention
Live courses work for activation energy — you’re excited, you’ve seen other people do it and you’re amazed by it, and “identity-wise, I feel like the sort of person that should be using code more.” But you can’t get over the hump alone. “I don’t think you can get over the hump very easily on your own”
The social element is irreplaceable — being in a room (virtual or physical) full of people all trying the same thing, with instructors who care about your outcomes, creates motivation that recordings can’t replicate
The sneaky benefit: The biggest complaint Taylor hears from consulting clients: “They don’t have time to play with this stuff — they don’t have time to save themselves time.” A full-day workshop buys you at least half a day of protected time to actually build. “Nobody’s going to bother you if you’re in a Claude Code camp all day”

The next cohort of Claude Code for Absolute Beginners, taught by Mike Taylor, takes place April 14, 2026. “Even if you don’t take the course, just download Claude Code and start messing around with it.”

Marq AI Wiki

Explorer

Claude Code for Absolute Beginners — What We've Learned Teaching AI to Non-Coders