Introduction to Programming for AI

SUMMARY

Eli Etherton, known as Eli the Computer Guy, delivers an in-person class on programming basics for AI projects, covering language selection, tools, architectures, project management, and practical AI integration like LLMs and agents.

STATEMENTS

Eli the Computer Guy has been educating on tech via YouTube since 2009 and founded Silicon Dojo to bring in-person tech education worldwide.
Silicon Dojo aims to make tech education accessible like YouTube videos but in physical spaces, countering the shift to online-only learning.
The main website silicondojo.com lists classes, uses Meetup for RSVPs, and plans Asheville sessions since Eli is from North Carolina.
In-person classes are recorded for later viewing on Silicon Dojo's site, including fireside chats with tech leaders like IBM's AI VP and cybersecurity CEOs.
Silicon Dojo uses Python as the primary language, Bottle for web apps, and SQLite for relational databases in classes.
Self-study programs on the site include multi-part Python series, SQLite, and intro to programming, free with membership for email updates.
Education is free to end users, funded by crowdfunding raising about $30,000 for operations like hotels and offices.
Crowdfunding allows voluntary support without tying resources to donations, maintaining accessibility.
Eli's old site elithecomputer guy.com hosts years of free classes from 2010 Linux videos to 2019 updates.
All class materials, including workbooks and code, are on GitHub at SiliconDojo for replication and teaching others.
Upcoming classes include Twilio and 11 Labs for voice AI on February 11th, full-day SQL and Bottle web apps, and networking.
Future full-day classes will charge $1 to ensure commitment via credit card entry, accommodating up to 100 but limiting in-person to 24-30.
AI will not kill programming; it's essential for tech professionals to learn coding basics in about 40 hours.
Coding differs from computer science, which involves complex math; basic scripting builds useful tools quickly.
Learning to code enables auditing AI outputs and understanding front-end, back-end, databases, and cloud options.
Vibe coding by non-technical leaders signals mismanagement; tech pros get paid for architecture and cost-effectiveness, not just coding.
Efficient models like IBM's Granite 350M run on minimal hardware, emphasizing cost-effective scaling.
Cloud services like AWS or Azure start free but scale expensively; local infrastructure can cap costs for high-usage scenarios.
Vendor lock-in risks arise from vibe-coded systems using proprietary infrastructure without auditability.
Legacy systems persist due to critical services; new tech must integrate without full rewrites.
Vibe coding multiplies productivity for pros but suits MVPs for startups to validate user needs quickly.
Learning to code should solve personal or organizational problems, not just seek jobs; continuous skill-building prevents obsolescence.
Recruiters check GitHub activity; post-bootcamp inactivity signals low interest.
Talk to authority figures about org problems like hiring shortages to identify AI-solvable issues.
Choose languages by demand and earnings; C# pays more than PHP despite similar learning time.
Geographic location influences needed skills; Research Triangle favors Red Hat over Ubuntu.
Certifications matter only if aligned with local hiring; some startups value portfolios over certs.
Do not marry a stack; skills like if-else and loops transfer across languages.
Avoid identity-tied stack debates; listen to advice matching your context.
Focus on one skill at a time; chasing AI trends leads to dead ends.
No right answer in tech; solutions depend on resources, legacy, and constraints.
In-person college aids exploration over four years; online suits known paths but risks endless credentialing.
Bootcamps accelerate but atrophy without jobs; ensure local demand for taught stacks.
Self-study excels with direction; isolated learning like solo C++ yields impractical knowledge.
Institutional knowledge prioritizes team maintainability over individual expertise.
Total cost of ownership includes maintenance; Windows eases admin despite licensing vs. Linux's variability.
Stakeholder buy-in requires trust; demonstrate value to secure ongoing resources.
Tech isn't always best; training may outperform systems for simple workflows.
Speed varies by need; Python suffices for most AI without C++'s raw performance.
Regulations like HIPAA, PCI, and data sovereignty demand local or obfuscated processing.
Waterfall suits defined projects with endpoints; Agile enables endless iteration but risks scope creep.
Professionals anticipate unasked questions; clear expectations prevent post-delivery disputes.
Programming languages target specifics: PHP for dynamic sites, Swift for iOS.
Interpreted languages like Python need runtime installation; versions matter for compatibility.
Compiled languages avoid interpreters but require porting per OS.
Runtimes like Java enable cross-platform but need installation.
Full-stack developers build complete products; specialists risk unemployment in downturns.
IDEs like VS Code standardize workflows; extensions enhance coding.
Frameworks extend languages: Django for Python web apps, React for JS UIs.
APIs connect to services cheaply; service-oriented architecture outsources non-core functions.
Databases: relational for accuracy (SQL), NoSQL for scalability (e.g., social media).
Edge computing pushes AI to devices, reducing bandwidth via JSON summaries.

IDEAS

Bringing tech education back to in-person mirrors the 2009 shift from physical to online, reversing isolation trends.
Free end-user education funded separately abstracts costs, empowering global access without financial barriers.
AI hype flips past "everyone code" to "no one code," ignoring auditing needs for tech literacy.
Vibe coding by CEOs erodes trust; pros earn via whiteboard architecture, not rote tasks.
Granite model's tater-tot efficiency democratizes AI for low-resource environments.
Video streaming costs explode on clouds; colo racks offer fixed bandwidth for massive scale.
Legacy crap persists because upgrades risk critical downtime, forcing coexistence.
Vibe coding prototypes validate startups faster than full builds, avoiding Steve Jobs-style assumptions.
Decade-long paychecks in tech account for boom-bust cycles; laid-off time builds skills.
GitHub silence post-bootcamp dooms hires; continuous projects signal passion.
Organizational problems like receptionist shortages yield AI automations via Twilio/11 Labs.
Earnings gap: C# doubles PHP pay for equal effort, rewarding demand alignment.
Ruby on Rails revival from unmaintained 2015 projects highlights cyclic tech demands.
East Coast PHP vs. West Coast Rails shows geography dictating stack preferences.
Certifications signal irrelevance without portfolios in startup-heavy areas.
Stacks evolve; Python today may fade, but core logic endures.
Me-bro culture ties identity to stacks, breeding toxic debates over practical advice.
Chasing AI trends like MCP leads to obsolescence; sequential learning builds depth.
College's four-year exploration beats rushed bootcamps for undecided learners.
Bootcamp acceleration fades yearly without refinement; local stack mismatch wastes investment.
Institutional knowledge ensures project survival post-heart attack, prioritizing team over ego.
Linux's "free" hides admin costs; Windows' ubiquity eases repairs.
PHP's cheap devs enable AI-integrated web apps via curl, balancing cost and function.
Buy-in from execs via trust turns "dumb" ideas into loyalty after failures.
Prompt engineering swaps vendor LLMs but breaks contexts, demanding redesign.
RAG externalizes knowledge to databases, letting LLMs handle style without retraining.
AI as "just another stack" demystifies hype, revealing API-heavy plumbing.
Quantized models retain vast knowledge on tiny hardware, trading precision for access.
Korean-only models minimize hallucinations by narrowing training scope.

INSIGHTS

Tech education thrives in hybrid physical-digital spaces, combating online fatigue while leveraging global reach.
Programming literacy audits AI outputs, preventing costly CTO blind spots in startups.
Cost-effective AI favors lightweight models on legacy hardware over flashy clouds for sustainable scaling.
Legacy integration demands coexistence strategies, as full rewrites ignore entrenched critical paths.
Vibe coding accelerates MVPs but risks unmaintainable spaghetti without architectural oversight.
Long-term career resilience requires problem-solving projects over job-chasing credentials.
Geographic and vertical alignment optimizes stack choices, turning local demands into high-earning niches.
Stack fluidity fosters adaptability; rigid identities stifle professional growth amid trends.
Sequential skill mastery prevents AI trend-chasing burnout, building transferable foundations.
Education paths must balance exploration with employability, avoiding credential traps.
Team maintainability trumps individual brilliance, ensuring institutional resilience.
Total ownership calculus favors familiar ecosystems over "free" alternatives with hidden expertise costs.
Stakeholder trust converts skepticism to advocacy, unlocking perpetual resources.
Non-tech solutions like training often outperform AI for workflow tweaks.
Prompt and context engineering treats LLMs as black boxes, emphasizing input precision over internals.
RAG decouples knowledge from models, enabling efficient, updatable expertise infusion.
AI agents simplify to API orchestrations, revealing "intelligence" as coordinated services.
Quantization unlocks elite models for edge devices, prioritizing deployment over perfection.

QUOTES

"What if we started doing uh tech education basically in the real world. What if we made tech education as accessible as a YouTube video but actually in person?"
"Free to the end user. So the idea is you abstract out the financial component from the education component."
"AI will not kill programming; it's essential for tech professionals to learn coding basics in about 40 hours."
"Our job as technology professionals is this thing. This is what we get paid for. We get paid for the whiteboard."
"It will run on a potato, but as I said, it will run on a tater tot."
"You start to get a couple of customers on the platform. It doesn't cost you much money. Here's the interesting thing. you start seeing the bills... once they start getting a thousand users or 10,000 users it's amazing how the price goes through the roof."
"If your CEO starts to vibe code... quit. Just walk the hell out the door."
"Vibe coding is a force multiplier for technology professionals."
"Think about your paycheck over something like a decade rather than on a month-to-month basis."
"Recruiters... would literally go to GitHub and... 90% of the people he knew the day that their boot camp graduated because you would see all of these entries to GitHub and then it would stop."
"You do not marry a stack, you marry a spouse."
"If all you have is your stack, there are a lot of people out there the only thing of value in their life is they went to a boot camp, they learned JavaScript and by God that's what makes them important on this planet."
"There is no right answer in technology."
"A professional answers the questions the customer does not know to ask."
"Programming languages are built to do specific things."
"All an IDE is is a program that is designed to actually write code."
"Frameworks can be exponentially more difficult to learn than the language itself."
"APIs allow you to get all the power of them spending a trillion dollars... for a buck 50 per million tokens."
"AI does not exist. AI is a marketing term for billionaires trying to become trillionaires."
"All AI is is it's just another stack."
"Hallucinations are just errors. Again, it's statistically trying to put things in order and it just simply puts incorrect data in order."

HABITS

Record all in-person classes for on-demand access and review.
Use Meetup for RSVPs while hosting core content on owned platforms like silicondojo.com.
Maintain self-study programs with 50+ part series on Python, SQLite, and programming fundamentals.
Share all class materials on GitHub for free replication by others.
Crowdfund operations transparently, avoiding donation-resource ties.
Schedule classes around personal life, announcing via email blasts to members.
Teach using Python, Bottle, and SQLite as standard stack for consistency.
Mentor founders by auditing code and advising on tech hires.
Build projects solving real problems, like automated receptionists.
Research local demands before certifying, e.g., Red Hat in Research Triangle.
Avoid stack fixation; learn transferable patterns like wireframing.
Listen to contextual advice, ignoring dogmatic stack debates.
Focus sequentially: master Python before databases, then AI integration.
Use VS Code as primary IDE for team standardization.
Copy-paste frameworks like Django to accelerate development.
Prototype iteratively with user feedback before full builds.
Manage expectations by clarifying AI limitations upfront.
Update classes quarterly to counter rapid AI changes.
Cache SQL queries to minimize LLM calls for efficiency.

FACTS

Eli's YouTube tech education since 2009 has empowered people from Italy to Grand Canyon.
Silicon Dojo crowdfunding totals $30,000 for operations.
PHP developers in Asheville earn $40,000-$50,000 annually.
AWS charges 9 cents per gigabyte for data transfer; 1-hour 1080p video is 3 gigs.
Colocation rack costs $1,000/month plus $200/gig in/out, supporting 200-300 concurrent 1080p streams.
OpenAI declares 2026 as "year of productivity" for useful AI applications.
Bootcamps cost up to $20,000, comparable to some college degrees.
Ruby on Rails jobs surged in 2022-2023 from unmaintained 2015 projects.
West of Mississippi favored Novell, east favored Microsoft in 2000.
MCSE certifications involved 30 years of Windows admin knowledge.
Granite 350M model runs on 2012 MacBook Pro with 4GB RAM.
Python 3.8+ required for many AI modules; older versions like 3.5 cause compatibility issues.
LAMP stack: Linux, Apache, MySQL, PHP; Windows alternative: IIS, MS SQL.
NoSQL enables social media scale but risks data loss (1/1000 posts acceptable).
OpenAI's GPT-4o mini: $0.15/million input tokens, $0.60/million output tokens.
1,000 tokens ≈ 750 words; images/voice use specialized token counts.
South Korean small model trained solely on Korean minimizes hallucinations.
Blues IoT cards: $60 for 10-year, 500MB data, JSON-only via satellite.
RAG chunks documents to 300-500 tokens for vector embedding.
OpenAI committed to $1.3 trillion in contracts.

REFERENCES

YouTube channel: Eli the Computer Guy (since 2009).
Silicon Dojo website: silicondojo.com (classes, fireside chats, self-study).
Meetup for RSVPs.
GitHub: github.com/SiliconDojo (In-Person classes, workbooks, code).
Donorbox: donorbox.org/etcg (crowdfunding).
LinkedIn: /eli-etherton-a15362211.
elithecomputerguy.com (old classes, Linux videos 2010-2019).
IBM Granite models (350M parameters, business-trained).
Python (primary language, 50-part series).
Bottle (web app framework).
SQLite (relational database).
Twilio (voice AI, SMS).
11 Labs (text-to-voice).
SQL (full-day class).
OpenAI (LLMs, Whisper, APIs).
RAG (retrieval augmented generation).
MCP (model context protocol).
VS Code (IDE).
Django (Python web framework).
React (JS frontend framework).
OpenCV (computer vision).
Tkinter (GUI framework).
Figma (wireframing software).
Blues (IoT cards, satellite JSON).
LLaVA (edge vision model).
Plato/Socrates (Socratic method for requirements).
Fire Ship (YouTube channel on databases).

HOW TO APPLY

Join Silicon Dojo membership for free access to self-study and email updates.
Attend in-person classes via Meetup RSVPs; watch recordings if missed.
Use GitHub repo for class workbooks and code; replicate for personal teaching.
Support via Donorbox if valuable, without expecting tied resources.
Explore elithecomputerguy.com for historical tech classes like 2010 Linux.
Schedule $1 commitment for future full-day classes to ensure attendance.
Dedicate 40 hours to Python basics for functional scripting ability.
Audit AI-generated code by learning front-end, back-end, and database differences.
Evaluate costs: compare cloud scaling (AWS/Azure) vs. local colo for high-usage.
Prototype MVPs with vibe coding to validate user interest before full investment.
Maintain GitHub activity post-learning to demonstrate ongoing commitment.
Interview bosses on org problems like hiring; brainstorm AI solutions.
Research local job demands via meetups before choosing languages/certifications.
Learn design patterns for cross-language reuse, like wireframing.
Select one skill (e.g., Python) and master before branching to databases.
Use VS Code with extensions for standardized coding environments.
Copy-paste frameworks like Django for efficient web app builds.
Draw wireframes on paper with stakeholders for UI buy-in.
Write pseudocode in human terms before translating to syntax.
Build prototypes for clickable validation; iterate via rapid prototyping.
Deploy MVPs to test free usage; adjust based on feedback.
Manage AI expectations: clarify it's pattern-matching, not true intelligence.
Use RAG for company docs: chunk, embed, query via vector DB.
Quantize models for edge deployment, accepting precision trade-offs.
Cache SQL results to reduce LLM calls for repeated queries.
Design agents as API orchestrators for tasks like scheduling.
Update AI stacks quarterly to handle rapid changes in modules/APIs.
Train users on keywords for simple if-else over complex LLMs.

ONE-SENTENCE TAKEAWAY

Master programming basics to architect cost-effective AI solutions solving real problems.

RECOMMENDATIONS

Prioritize Python for AI due to its simplicity and ecosystem dominance.
Invest 40 hours in core coding to enable quick, useful scripting.
Audit all AI outputs personally to avoid costly implementation errors.
Choose lightweight models like Granite for resource-constrained environments.
Opt for local infrastructure in high-scale scenarios to cap bandwidth costs.
Flee vibe-coding leadership as it undermines technical integrity.
Build MVPs rapidly with AI tools to validate market fit early.
Maintain GitHub portfolios with continuous projects for recruiter appeal.
Consult superiors on org pain points to target impactful AI applications.
Align stacks with local geography and verticals for optimal employability.
Pursue portfolios over certifications in startup ecosystems.
Embrace stack flexibility; focus on transferable logic like loops.
Seek advice from listeners who consider your full context.
Sequence learning: Python first, then databases, then AI integration.
Explore in-person college for undecided paths; self-study for directed goals.
Evaluate bootcamps by local stack demand and post-grad job acceleration.
Design for institutional knowledge; ensure team maintainability.
Calculate total ownership: weigh licensing against admin expertise.
Secure buy-in by building trust through honest failure predictions.
Question if training beats AI for simple workflow improvements.
Use prompt engineering judiciously; test across LLM vendors.
Implement RAG for knowledge bases to avoid model retraining.
Treat AI as a stack: APIs, prompts, contexts form the core.
Quantize aggressively for edge AI to minimize hardware needs.
Cache queries in SQL hybrids to economize on LLM usage.
Prototype agents as simple API chains before overcomplicating.
Update AI dependencies quarterly to preempt breakage.

MEMO

Eli Etherton, better known as Eli the Computer Guy, stood before a room of eager learners in Research Triangle, North Carolina, launching his Silicon Dojo initiative with a sprawling two-hour-plus seminar on "Introduction to Programming for AI." A veteran YouTuber since 2009, Etherton has long demystified tech for global audiences—from Italian enthusiasts to Grand Canyon trekkers. Now, he's flipping the script on post-pandemic isolation, insisting that in-person education recaptures the tactile spark missing from endless Zoom calls. "What if tech classes felt like YouTube but with real faces?" he quipped, unveiling free resources on silicondojo.com, GitHub repos ripe for replication, and a crowdfunding model that's raised $30,000 without nickel-and-diming attendees. His ethos: education abstracted from finance, empowering anyone to teach or tinker.

Dispelling AI doomsayers, Etherton argued that far from obsoleting coders, large language models demand sharper auditing skills. "Kids should not learn to program? That's stupid," he declared, likening basic scripting to essential literacy—achievable in 40 hours, distinct from math-heavy computer science. He skewered "vibe coding" executives who fancy themselves developers, warning of vendor lock-in pitfalls and legacy system traps that haunt enterprises. Drawing from mentorship mishaps—like a CTO unable to audit $300,000 in PHP blunders—Etherton stressed the whiteboard's true value: architecting cost-effective stacks. Granite's 350-million-parameter model, runnable on a "tater tot," exemplifies efficient scaling, while cloud bills exploding at 1,000 users underscore local infrastructure's edge for video-heavy apps.

Etherton's pragmatic lens extended to career navigation in tech's boom-bust cycles. Forget monthly paychecks; view earnings over decades, he advised, urging continuous GitHub commits to signal passion beyond bootcamps. Tailor stacks to locales—Red Hat in Research Triangle, PHP on the East Coast—and shun "marrying" languages, as skills like if-else transcend syntax. He critiqued bootcamps' $20,000 price tags for atrophy risks and championed full-stack versatility amid layoffs, where frontend wizards struggle without backend heft. Project management wisdom followed: Waterfall for defined endpoints, Agile for iterative behemoths like social media, but always anticipate unasked questions to manage expectations.

Diving into AI's nuts and bolts, Etherton peeled back the hype: LLMs are "marketing for billionaires," pattern-matchers prone to Reddit-fueled biases, not sentient oracles. Prompt engineering and RAG—chunking docs into vector databases—unlock practical power without retraining models, while agents boil down to API orchestrations for tasks like scheduling. He favored open APIs for trillion-dollar infrastructure at pennies per token, cautioning against overkill like $50,000 Mac clusters when a $150 Raspberry Pi processes images overnight. Regulations loom large—HIPAA, data sovereignty—pushing obfuscation or local runs. Ultimately, Etherton framed AI as "just another stack," a reminder that beneath the glamour lies code, context, and conversation—tools for human flourishing in a machine-augmented world.

As the class wrapped, attendees grappled with rapid evolution: quarterly updates now mandatory, lest modules break legacy builds. Etherton's Silicon Dojo, with its $1 commitment classes and replicable ethos, promises a dojo for the digital age—where novices wireframe dreams, pros debug realities, and AI serves problems, not myths. In an era of trillion-parameter promises, his message resonates: Start small, think big, code wisely.