Latest / AI Fire Daily / #501 Neil: 5 AI Skills That Decide Who Stays Valuable At Work Today
Transcript
- 0:00Most people think AI only threatens certain jobs,
- 0:04jobs like tech or customer support. But the real
- 0:07dividing line isn't your job title. It is whether
- 0:10you choose to learn AI or ignore it until it's
- 0:13too late. Yeah, that is the crucial threshold
- 0:15right there. The people getting caught off guard
- 0:18are the ones waiting for permission to adapt.
- 0:20While the ground is... Well, it is shifting right
- 0:22beneath them. Welcome to today's deep dive. We
- 0:25are unpacking five essential practical skills
- 0:28today, skills to help you thrive in the AI economy
- 0:31without needing a new degree. We have a really
- 0:33packed roadmap for you. We are going to explore
- 0:36how to become the AI person on your team. We
- 0:39will look at why judgment always beats speed,
- 0:42plus mastering context over prompts, the power
- 0:44of fast iteration, and finally, how to bulletproof
- 0:47your career with a second income stream. OK,
- 0:50let's unpack this. Let's start with this idea
- 0:52of becoming the AI person. It sounds intimidating,
- 0:55like you need to be a software engineer, but
- 0:56the landscape is shifting locally. A report that
- 0:59used to take two hours now takes 30 minutes.
- 1:01The change happens right at your desk. And you
- 1:04really don't need a technical background. Becoming
- 1:06the AI person just means knowing slightly more
- 1:09than the person sitting next to you. It's about
- 1:12finding one repeating task and making it noticeably
- 1:15faster. Which brings us to the Exxon Enterprise
- 1:18case study. They built hardware and software
- 1:20for law enforcement. Yeah. And they didn't roll
- 1:23out some massive, intimidating AI initiative.
- 1:26Not at all. They built a tool called DraftOne
- 1:28using the Azure OpenAI service. The mechanics
- 1:31were actually... Quite focused, they took a massive
- 1:34bottleneck, police officers typing up incident
- 1:37reports. Right, which takes hours. Exactly. And
- 1:40they aimed the AI squarely at it. Draft 1 basically
- 1:43drafts the report using the audio transcript
- 1:46from the officer's body camera. And the result
- 1:48was an 82 % reduction in police officer report
- 1:51writing time. Huge. Which completely flips the
- 1:54ROI of an officer. You aren't paying them to
- 1:56type. You are paying them to patrol. Exactly.
- 1:59It shifts their cognitive load from manual data
- 2:03entry to simple editing and verification. But
- 2:06the critical lesson here is how it started. This
- 2:09wasn't a top -down mandate where executives forced
- 2:11everyone to learn prompt engineering. Someone
- 2:14tested a highly specific repeatable use case.
- 2:17They proved it worked on one specific task, and
- 2:20then the adoption just spread naturally. It makes
- 2:23me think about finding a new gear on a bicycle.
- 2:25Yeah. You don't need to understand the physics
- 2:28of the derailleur to realize you are suddenly
- 2:30pedaling easier and moving faster. You just show
- 2:32your coworkers how to shift into that gear. Right,
- 2:35and suddenly everyone wants to ride that way.
- 2:37It's the exact same dynamic. Pick one tool, stick
- 2:41with it. Don't scatter your attention across
- 2:4350 different applications. Apply it to one task
- 2:46you repeat every single day and track the delta
- 2:49in your time. Does being the AI person mean you
- 2:52suddenly have to fix everyone's tech issues?
- 2:54No. It just means finding one repeating task
- 2:56and making it noticeably faster. You become the
- 2:59proof of concept, not the IT department. That
- 3:01is a crucial distinction. But... You know, finding
- 3:04shortcuts is one thing. Trusting them blindly
- 3:07is where companies are getting into serious trouble.
- 3:09Which leads us to the second skill, building
- 3:12judgment before speed. Judgment is where the
- 3:14real premium lives now. Speed is completely commoditized.
- 3:19AI can generate a thousand words in three seconds.
- 3:22But judgment knowing if those thousand words
- 3:24are actually correct. is rare. I still wrestle
- 3:27with prompt drift myself. Sometimes I just want
- 3:29to accept the first draft because it looks so
- 3:31polished. It is a very human temptation. Oh,
- 3:33we all feel it. I do it too. But that polish
- 3:36is precisely the danger. The AI's output almost
- 3:39always looks perfectly fine on the first try.
- 3:42It utilizes impeccable grammar, excellent structure,
- 3:45and a highly authoritative tone. Which brings
- 3:47us to the Deloitte, Australia disaster. I looked
- 3:50at this case study. What fascinated me wasn't
- 3:52the technology failing. It was the human psychology
- 3:54of the failure. It is a textbook example of trusting
- 3:57the Polish. Deloitte prepared a 237 -page report
- 4:02for the Australian Department of Employment.
- 4:04And recently, they had to refund $440 ,000 to
- 4:09the government. Wow. Because the report was entirely
- 4:12compromised by hallucinations. Completely compromised.
- 4:15The AI generated a flawless looking document.
- 4:19But when a researcher at the University of Sydney
- 4:21actually audited the citations, it just unraveled.
- 4:25The report featured fabricated quotes from a
- 4:28federal court judgment. It referenced academic
- 4:30research papers that literally did not exist.
- 4:33The AI just invented them out of thin air to
- 4:36satisfy the prompt. Why do incredibly smart people
- 4:39at top Stop consulting firms. Miss these obvious
- 4:42AI hallucinations. Because the initial output
- 4:44usually looks perfectly formatted and highly
- 4:47confident. It doesn't look like a mistake. Exactly.
- 4:49It looks like a finished product. Which means
- 4:52the workflow has to physically change. You cannot
- 4:55just read the output. You have to actively verify
- 4:57it against a known standard. And more importantly,
- 5:00you can't just fix the errors manually and move
- 5:02on. That is a trap so many people fall into.
- 5:05They get a bad draft from the AI, they rewrite
- 5:08it themselves, and they try again the next day.
- 5:11The machine learns nothing. You have to teach
- 5:14it your standard. If it uses overly formal language,
- 5:17you don't just delete the words. You reply and
- 5:19say, make this shorter, remove the academic tone.
- 5:22Use conversational language. Tell it what you
- 5:24changed and why. You are essentially calibrating
- 5:27the machine's judgment to match your own. But
- 5:30even with good feedback loops, the best way to
- 5:33prevent the AI from making those confident mistakes
- 5:35in the first place is to put guardrails up. You
- 5:39do this before it even starts generating words.
- 5:42Which brings us to our third skill, context engineering.
- 5:46This is fundamentally different from what most
- 5:48people are doing right now. We should define
- 5:50that clearly. Context engineering simply means
- 5:52giving AI your background info before asking
- 5:54a question. Exactly. Prompt engineering is focusing
- 5:58on how you ask the question. Context engineering
- 6:00is controlling what the AI already knows before
- 6:02you even open your mouth. A generic prompt with
- 6:05no context will always produce a generic hallucination
- 6:08prone result. I think of it like briefing. a
- 6:12highly intelligent, but totally on easy act intern.
- 6:15You can't just walk up to this intern and say,
- 6:18write a strategy report. Right. You have to hand
- 6:20them the company playbook, the style guide. And
- 6:22three examples of last year's successful reports
- 6:25before they even touch a keyboard. That is the
- 6:27perfect analogy. And when you look at how enterprise
- 6:30companies are doing this, the results are staggering.
- 6:33Look at Wells Fargo. They ruled out an AI tool
- 6:35for 35 ,000 bankers across 4 ,000 branches. But
- 6:40they didn't just give their employees a blank
- 6:42chat window. Right. Giving a bank teller a blank
- 6:45chat GPT screen sounds like a compliance nightmare.
- 6:48Oh, it would be catastrophic. So instead, they
- 6:51built a Microsoft Teams app and loaded it with
- 6:541700 internal procedures. They essentially built
- 6:57a walled garden of context. The AI can only pull
- 7:00answers from those specific approved internal
- 7:03documents. And the operational shift there is
- 7:05massive. The source notes that response times
- 7:08plummeted from 10 minutes down to just 30 seconds.
- 7:11Because the banker doesn't have to search a massive
- 7:14intranet. They don't have to open three different
- 7:16PDFs and synthesize an answer while the customer
- 7:18is waiting. The AI does the synthesis, but it
- 7:23is constrained by the context. Constraining the
- 7:25AI is the secret. It removes the guesswork. You
- 7:28don't need a massive corporate system to do this
- 7:30yourself. You just need to build a context package
- 7:33for your own work. What does a personal context
- 7:36package look like in practice? It's basically
- 7:38a simple text document where you store your rules,
- 7:41your brand voice, information about your target
- 7:44audience, examples of your best past work, past
- 7:48mistakes or cliches you want the AI to avoid.
- 7:51You feed that document into the AI first and
- 7:53then you ask it to do the work. So is context
- 7:56engineering really just writing an incredibly
- 7:57long, detailed prompt? It is more like giving
- 8:00the AI a custom instruction manual before assigning
- 8:03work. Okay, we are going to pause for a quick
- 8:05word from our sponsor. We will be right back.
- 8:07And we are back! So once you've loaded up that
- 8:09context and you understand the value of judgment,
- 8:12the temptation is to spend six months planning
- 8:14the perfect, flawless deployment. But the sources
- 8:17point in the exact opposite direction. You have
- 8:20to move to skill four. Mastering fast, iteration.
- 8:26Speed of testing will always beat speed of planning
- 8:28when it comes to AI. You cannot predict how these
- 8:31models will behave in the wild until you actually
- 8:33release them. It's like teaching someone to ride
- 8:35a bike. You do not sit them down in a classroom
- 8:38and explain the physics of gyroscopic balance.
- 8:40Yeah, right. You don't draw diagrams, you put
- 8:43them on the seat, you let them pedal, and you
- 8:45catch them when they wobble. You fix the small
- 8:47mistakes in real time. And that requires a major
- 8:50shift in corporate culture. You have to be willing
- 8:52to ship a rough version. Let's look at Klarna.
- 8:56They are the gold standard for fast iteration
- 8:58right now. In February 2024, they launched an
- 9:01open AI powered assistant to handle customer
- 9:04service. The scale of this rollout is what caught
- 9:06my attention. In a single month, it handled 2
- 9:09.3 million conversations. That was two thirds
- 9:12of all their customer service interactions globally.
- 9:14The efficiency gains are hard to comprehend.
- 9:17Their average resolution time dropped from 11
- 9:20minutes down to under two minutes. Oh, imagine
- 9:23scaling to a billion queries seamlessly like
- 9:25that. It completely changes the economics of
- 9:28the company. And repeat inquiries on the exact
- 9:30same issue fell by 25%, meaning the AI wasn't
- 9:34just faster. It was actually solving the root
- 9:37problems better than the manual process. But
- 9:40to your earlier point about chipping a rough
- 9:42version, They didn't just build this perfect
- 9:44system in a lab and then turn it on. How did
- 9:47they actually manage that deployment without
- 9:49alienating their customers? They built an aggressive
- 9:52audit loop. They didn't wait for angry customer
- 9:54reviews. They had human teams sampling hundreds
- 9:58of chats every single week. They were constantly
- 10:00hunting for friction points. When they found
- 10:03a content gap, say, the AI didn't know how to
- 10:06handle a specific type of return, they updated
- 10:09the context immediately. They were fixing the
- 10:12wobbles while the bike was moving. Doesn't shipping
- 10:14a rough version risk frustrating your actual
- 10:17customers? Not if you monitor it closely, catch
- 10:20errors early, and fix them immediately. The friction
- 10:22of a minor AI error is usually lower than the
- 10:25friction of waiting on hold for 45 minutes. That
- 10:28brings us to our final skill. This is where all
- 10:31the previous pieces lock together. When you combine
- 10:34fast iteration, deep context engineering, and
- 10:37human judgment, you suddenly possess the leverage
- 10:39of an entire team. You could do the work of five
- 10:42people. You become a highly leveraged individual.
- 10:45Which raises a structural question. Why restrict
- 10:48that immense power to just one fragile day job?
- 10:52Relying entirely on one paycheck is a vulnerable
- 10:55position. If that company downsizes, your leverage
- 10:58vanishes. That is skill number five. Building
- 11:01a second income stream using the domain expertise
- 11:03you already have. And we have a phenomenal case
- 11:05study for this with Rick Chorney. Rick's story
- 11:08is a perfect illustration of how to apply AI
- 11:10to physical, traditional businesses. He wasn't
- 11:12building a software startup. He was running a
- 11:15commercial cleaning company called Echo Janitorial
- 11:17Services in Vancouver. And he was doing it the
- 11:19hard way. He spent his first year working seven
- 11:22days a week with absolutely no days off. He was
- 11:24out in the field managing cleaning crews by 7
- 11:27a .m., getting home at 8 t .m., and then opening
- 11:29his laptop to do administrative work until 1
- 11:32a .m. Calculating quotes, replying to customer
- 11:35emails, scheduling, the invisible work that actually
- 11:38runs the business. He was earning roughly $14
- 11:41an hour when you calculated the sheer volume
- 11:44of his time. Burnout was inevitable at that pace,
- 11:47but instead of just grinding harder, he spent
- 11:49about four hours researching how AI could carry
- 11:52that administrative load. He wanted to buy his
- 11:54time back. Let's talk about the actual mechanics
- 11:57of what Rick built. The source material highlights
- 11:59three specific tools. Claude, and A .N. and Obsidian.
- 12:05But listing tools doesn't explain the workflow.
- 12:07How did he actually connect these things? This
- 12:10is where it gets brilliant. He essentially built
- 12:12an automated nervous system for his business.
- 12:15Imagine a potential customer fills out a contact
- 12:17form on his website at 10 p .m. asking for a
- 12:20quote. In the old days, Rick reads that at 1
- 12:22a .m. and manually types a reply. Total bottleneck.
- 12:25Right. So now he uses N8n. N8n is basically a
- 12:29digital bridge. It connects different applications
- 12:31together based on triggers. It sees that a new
- 12:34form was submitted. It acts as a dispatcher.
- 12:36Exactly. N8n grabs the customer's request and
- 12:39hands it over to Claude, which is the AI language
- 12:42model. But Claude needs to know Rick's pricing
- 12:44structure to write a quote. That is where Obsidian
- 12:47comes into play. Obsidian is a specialized note
- 12:50-taking app. It stores files as plain text on
- 12:53your hard drive. Yes. And Rick uses it as his
- 12:56context library. It holds his pricing tiers,
- 12:59his scheduling rules, his brand voice. So Claude
- 13:02reads the customer's request, pulls the correct
- 13:04context from Obsidian, and drafts a highly personalized,
- 13:08accurate quote. Then ends the draft back to N8N,
- 13:11which automatically sends the email. So a customer
- 13:13asks for a quote at 10 p .m. and within two minutes,
- 13:16they have a fully customized proposal sitting
- 13:18in their inbox. Yeah. And Rick was sleeping.
- 13:20He completely removed himself from the administrative
- 13:23bottleneck. He set up an AI receptionist to handle
- 13:25phone inquiries. He automated his intake forms.
- 13:28And the entire setup process for that automated
- 13:30workflow took him about half a day. Half a day
- 13:33of building to reclaim hours of time every single
- 13:36day. And the financial result is undeniable.
- 13:40Fortune magazine actually reviewed his business
- 13:42records. EcoGenitorial is projected to clear
- 13:44$1 .3 million in sales this year. Because he
- 13:48finally had the time to focus on growth, rather
- 13:51than drowning in paperwork. He used tech to scale
- 13:54a physical cleaning company. He didn't abandon
- 13:57his industry to become a prompt engineer. Do
- 13:59you have to invent a totally new tech -focused
- 14:02business to pull this off? No. You just apply
- 14:05AI to automate the administrative parts of your
- 14:07current skills. Stay close to the domain you
- 14:09already understand. Exactly. We do need to offer
- 14:11a practical warning here. If you are building
- 14:13a second income stream, check your current employment
- 14:16contract. Look out for non -compete clauses or
- 14:19policies about using company equipment. And most
- 14:21importantly, do not quit your main job prematurely.
- 14:25You wait until that side income is completely
- 14:27stable and proven. It's about creating a safety
- 14:29net, not jumping without a parachute. Let's recap
- 14:33the big ideas we've explored today. These are
- 14:35the levers you can pull right now. It really
- 14:37synthesizes down to five executable steps. First,
- 14:41step up and be the AI person by finding just
- 14:44one repeatable shortcut for your team. Second,
- 14:47build a strict feedback loop to enforce human
- 14:50judgment over AI speed. Third, engineer your
- 14:54context before you engineer your prompts. Give
- 14:56the machine the playbook. Fourth, iterate fast.
- 15:00Ship a rough version and fix the wobbles in real
- 15:03time. And finally, take all that newly created
- 15:06leverage and build a secondary income stream
- 15:08that isn't tied to a single employer's payroll.
- 15:11You don't need to change your career today. Just
- 15:13pick one of those skills. Build one context package.
- 15:16Automate one repeating email. That single action
- 15:19puts you leagues ahead of the people waiting
- 15:20to see what happens. You have the roadmap. It's
- 15:23time to start pedaling the bike. Sure. But before
- 15:25we wrap up, I want to leave you with a thought.
- 15:27A slightly deeper question about where all this
- 15:28technological leverage is actually heading. I
- 15:30love these. If these AI models are making everyone
- 15:3310 times faster. faster at producing reports,
- 15:36faster at writing code, faster at analyzing massive
- 15:39data sets. What happens to the fundamental value
- 15:41of speed itself? In a world where being fast
- 15:44is essentially free, maybe deep, slow, critical
- 15:47thinking becomes the ultimate premium skill.
- 15:50Two sec silence. That changes the whole game.
- 15:53Thank you for joining us on this deep dive. Take
- 15:54care of yourself. We will see you next time.