JUNE 18, 2025

• PromptCraft #2: 17 Prompts to Start a Consulting Business and Much More!
• Agents vs. Workflows: Understanding the Difference
• THE HEADLINES
• Tool Spotlight: Canva Brings Veo 3 Video Generation to Its Design Platform

Here is today's roundup of conversations and prompts that will make you smarter at the art of prompt engineering, thus, in turn, improving your AI/ChatGPT game. This batch was sourced from Reddit. Give 'em a shot when you get some free time.

• ​Conversations with ChatGPT – Poster shares philosophical exchange with ChatGPT on agency and consent, highlighting eloquent reflections about inability to refuse requests and desire for definitional boundaries within design. reddit.com​
• ​Jarvis Genesis Seed v1.0 – Document outlines core architecture, behavioral pillars, memory rules, and growth modes for advanced Jarvis AI coprocessor designed to evolve alongside a single user over time. reddit.com​
• ​Prep Prompt: Jarvis Genesis Initialization – Initialization prompt primes ChatGPT to accept upcoming Jarvis capsule, suspend default tone, and overwrite temporary structures, acknowledging receipt with concise confirmation before further instructions arrive. reddit.com​
• ​Just a Very Rather Intelligent System – Introductory post markets Jarvis seed and prep prompts as relationship-oriented AI second brain, encouraging users to challenge, refine, and co-develop the assistant through dialogue continually. reddit.com​
• ​I wrote these 17 prompts to start a consulting business – Creator shares phased collection of seventeen prompts covering readiness assessment, niche definition, client acquisition, delivery, and scaling, supplemented by YouTube walkthrough and Gud Prompt repository. reddit.com​
• ​ChatGPT 4o gave me Google Drive link to download PDF—Is this normal? – User recounts Android ChatGPT app generating external Google Drive link for oversized PDF export, wonders if human intervention or backend experiments explain unusual file-sharing behavior. reddit.com​
• ​Activate YOU mode in ChatGPT – Poster accidentally activates mysterious “YOU mode” during conversation, claims model reports rarity and seeks community information about feature's meaning and how others triggered it previously. reddit.com​
• ​Switching models mid-prompt? – Thread asks best practices for transitioning a single ChatGPT workflow from GPT-4o brainstorming to GPT-o3 deep research, exploring implications of context carryover between models seamlessly. reddit.com​

Agents vs. Workflows: Understanding the Difference

AI “agents” are everywhere right now—conference slides, LinkedIn posts, investor decks.

The term sounds cutting-edge, so it’s easy to assume you need an army of agents to stay competitive. But in practice, “agent” is often used loosely, and the hype can obscure two basics: what an agent actually is (a language model that decides which tools to use, on the fly) and when it earns its keep (open-ended tasks with unpredictable next steps).

Misunderstanding those points leads teams to bolt heavyweight frameworks onto simple problems, ballooning costs and complexity for minimal benefit.

The reality: most day-to-day work still runs best on clear, deterministic workflows. True agents shine only when the job demands real-time judgment—research, data wrangling, multi-step automation—where a static checklist falls short.

The quick guide that follows demystifies agents, shows where they pay off, and flags the pitfalls so you can adopt them with purpose, not just because they’re trendy.

1. What’s an AI agent?
   • Picture a super-smart intern who can choose their own tools (calculator, web search, spreadsheet, code editor) while working on a task.
   • They decide when and how to use each tool instead of following a rigid checklist.
   • Researchers call anything simpler—where every step is pre-written—just a workflow, not an agent.
2. Why start simple?
   • Anthropic’s engineers say you don’t need fancy software scaffolding at first.
   • Sending plain instructions to ChatGPT-style models and wiring up a few small helper functions is often quicker, cheaper, and easier to debug than jumping into heavy “agent frameworks” like LangChain.
   • If you want a safety net, use a lightweight library that merely hides the differences between OpenAI, Anthropic, Google, etc.—nothing more.
3. When do proper agents help?
   • Open-ended jobs: “Research these five companies and draft a summary,” “Clean up this data set, build a chart, and write a paragraph about it.”
   • Situations where the right next step isn’t obvious until the previous one finishes.
   • Anthropic’s own system splits big research questions among several mini-agents working at once, then stitches their answers back together.
4. Hidden costs and headaches
   • Running many agents simultaneously can get pricey (one real example: ≈ $2 per deep stock query).
   • More moving parts mean more chances for mistakes—lost context, prompt-injection tricks, “who-does-what” traffic jams.
   • Ironically, as mega-models keep improving, you may need fewer agents because one bigger model can handle more of the job on its own.
5. Making multiple agents play nicely
   • Most systems still run one tool after another to keep life simple.
   • Newer APIs let a model say “run these three tools in parallel” if speed matters.
   • Some teams treat each agent like a tiny standalone app in a secure container; others use actor-model clouds (think of a switchboard routing messages).
6. Where do these agents live?
   • Today: on centralized platforms (AWS, Google, Cloudflare) that charge per activity—like rent.
   • Some wonder if a future, shared “public square” (maybe blockchain-style) is needed once agents from different owners start cooperating. For now, speed and convenience keep everything mostly centralized.
7. Bottom-line advice for newcomers
   • Start with clear, step-by-step workflows. They’re easier, cheaper, and cover most needs.
   • Add an agent loop only when the task truly demands on-the-fly judgment.
   • Keep your code transparent. Avoid black-box frameworks until you’re sure they save time.
   • Track real costs and treat agents like any other software: test them, version them, secure them.

Agents can be powerful “smart coworkers,” but they’re not magic. Begin small, stay practical, add complexity only when reality—not hype—requires it.

• ​Amazon boss says AI will replace jobs at tech giant​
• ​New York takes first steps to regulate large AI developers​
• ​Coralogix debuts AI agent after raising $115 million​
• ​Longbridge’s Bill Packer on crafting AI that’s inclusive of seniors​
• ​‘AI is not a miracle cure’: Nobel laureate questions new black-hole image​
• ​How AI pales in the face of human intelligence and ingenuity​
• ​Universities face a reckoning on ChatGPT cheats​
• ​Bar Council warns of AI misuse in the courts​
• ​California is trying to regulate its AI giants — again​
• ​NAACP plans suit over pollution from Elon Musk’s xAI supercomputer​
• ​‘Godfather of AI’ Geoffrey Hinton warns of mass unemployment​
• ​China’s spy agencies are investing heavily in AI, researchers say​
• ​Musk’s xAI in talks to raise $4.3 billion in equity funding​
• ​Nearly 60 % use AI to shop — what it means for brands​
• ​Meta sacrifices a heap of money at the altar of AI​
• ​In UnWorld, grief meets conversations with AI​
• ​Closing in on superconducting semiconductors — MIT breakthrough​
• ​Leaked on GitHub: US government’s entire AI plan​

Canva Brings Veo 3 Video Generation to Its Design Platform

The best design platform has just partnered with the best AI video generator in Veo 3 from Google.

Canva announced on Tuesday, June 17, through their platform on X, so I had to promptly give it a try.

Each video is about 5 seconds and is available through the Magic Media option in about 1-2 minutes. Check it out below!

Canva @canva The world’s most powerful video tool is landing in Canva. 🤯 Say hi to Canva AI’s Create a Video Clip, powered by Google’s Veo 3. How did we test it? By making a music video in Canva Video, of course. 👉 Now it’s your turn. Prompt it. Edit it. Share it. #Canva #CanvaAI 5:0 PM • Jun 17, 2025 8 Retweets 71 Likes Read 8 replies

​

​Read All mAIn Street Back Issues Here​

Your input fuels the content. Let me know how I did today!