Learning Library

← Back to Library

Specific AI Career Path Strategies

Key Points

  • The usual “learn Python in 30 days” or “get a PhD/start a startup” advice is too generic, so you need concrete, role‑specific guidance to break into AI.
  • By 2030 AI is projected to add 170 million jobs but also wipe out 92 million, meaning entry‑level positions that traditionally serve as footholds are disappearing.
  • Research shows a noticeable drop in employment for young workers (22‑25) in AI‑exposed roles, highlighting the paradox of simultaneous opportunity and automation.
  • AI should be viewed as a diverse “career maze” with many distinct paths—much like the various roles in the food industry—rather than a single monolithic field.
  • The speaker analyzed 17 current AI career tracks, mapped hiring trends, and offered prompting strategies to help individuals identify and scale the specific AI role that fits their background.

Full Transcript

# Specific AI Career Path Strategies **Source:** [https://www.youtube.com/watch?v=Z88hYmdxp00](https://www.youtube.com/watch?v=Z88hYmdxp00) **Duration:** 00:17:08 ## Summary - The usual “learn Python in 30 days” or “get a PhD/start a startup” advice is too generic, so you need concrete, role‑specific guidance to break into AI. - By 2030 AI is projected to add 170 million jobs but also wipe out 92 million, meaning entry‑level positions that traditionally serve as footholds are disappearing. - Research shows a noticeable drop in employment for young workers (22‑25) in AI‑exposed roles, highlighting the paradox of simultaneous opportunity and automation. - AI should be viewed as a diverse “career maze” with many distinct paths—much like the various roles in the food industry—rather than a single monolithic field. - The speaker analyzed 17 current AI career tracks, mapped hiring trends, and offered prompting strategies to help individuals identify and scale the specific AI role that fits their background. ## Sections - [00:00:00](https://www.youtube.com/watch?v=Z88hYmdxp00&t=0s) **Navigating AI Career Paradoxes** - The speaker critiques vague AI‑break‑in advice, highlights data showing both massive job creation and loss, and urges concrete, role‑specific strategies to seize emerging AI opportunities while avoiding entry‑level automation. - [00:04:56](https://www.youtube.com/watch?v=Z88hYmdxp00&t=296s) **Designing Predictive Skill Assessment Prompts** - The speaker explains how they craft interview‑style prompts that anticipate meaningful signals—such as production deployments for ML engineers or editorial editing for prompt engineers—to accurately gauge candidates’ true AI and engineering proficiency. - [00:08:25](https://www.youtube.com/watch?v=Z88hYmdxp00&t=505s) **Evolving Product Management with AI** - The speaker outlines a method for merging traditional product management competencies with AI expertise, using an LLM‑driven assessment to generate a personalized, dynamically updated learning roadmap. - [00:11:59](https://www.youtube.com/watch?v=Z88hYmdxp00&t=719s) **Upstream Prediction for Career Success** - The speaker explains an AI‑career assessment that pre‑emptively flags common pitfalls and leverages early‑stage predictive insight—like meme‑driven probability forecasts—to guide users quickly toward viable roles, avoiding costly trial‑and‑error. - [00:15:19](https://www.youtube.com/watch?v=Z88hYmdxp00&t=919s) **Predictive AI Career Pathways** - The speaker argues that the exploding AI job market demands a personalized, data‑driven approach where carefully crafted prompts guide LLMs to predict an individual’s optimal career move by accounting for market timing, transferable skills, and common hiring blind spots. ## Full Transcript
0:00Most of what you read about how to break 0:02into AI is wrong. Not because it's 0:05incorrect, but because it's too vague to 0:07be useful. It's too high level. It's too 0:10general. Learn AI. Don't do that. Get 0:12specific. Get useful. I have spent the 0:15last few days taking apart 17 different 0:18career paths in AI that people are 0:20hiring for right now. I've looked into 0:22the job market, both where we're hiring 0:25and also where jobs are being destroyed. 0:27And I've come out of all of it with some 0:29overall takeaways for you as well as 0:31some approaches to prompting that help 0:33you to figure out where you can career 0:36path yourselves and where you can scale 0:38up if you have a particular job role in 0:40mind. So let's dive into it. First up, 0:43let's get at the overall job situation. 0:46The World Economic Forum, I know you 0:48love to read them, says 170 million new 0:50AI jobs will be created by 2030. Yay, 0:52yay. Break out the party hats. But 92 0:55million jobs will be destroyed in the 0:57same period. So net gain just 78 million 1:00positions projected. Stanford published 1:03research showing 13% unemploy or 1:05employment decline for workers aged 22 1:07to 25 and AI exposed roles since late 1:102022. That is not news to anybody I know 1:12who is in that age demographic. Look at 1:14my gray hairs. I am not. So AI is 1:16simultaneously creating opportunities 1:18and it's automating entry-level 1:20positions people use to break in. That's 1:23the paradox that we need to wrestle with 1:24and solve. So when you search how to 1:26break into AI in that world, you get one 1:29of two extremes. You get let's learn 1:31Python in 30 days and become an ML 1:34engineer. Lots of lots of that. Or you 1:37get a little bit of gatekeeping like you 1:38need to be a PhD from Stanford or you 1:40need to get into Y Combinator, you need 1:41to found your own startup. Both of these 1:44are suboptimal because they only apply 1:47to a very narrow subsegment of the 1:49overall population. If you look at the 1:51overall number of people interested in 1:53AI roles, not very many of them are 1:55going to be learning Python in 30 days. 1:57If you look at the overall number of 1:58people interested in AI, not very many 2:00of them are getting into Y Combinator 2:02and founding a company. And nobody talks 2:04about it as if AI is not a single career 2:06path but a but a wild maze of career 2:09paths that's developing. And I want to 2:11start talking about it like that because 2:13that's how it actually is. Right now, 2:15when people say, "How do I break into 2:17AI?" It's sort of like a chef hearing, 2:19"How do I break into food?" I mean, 2:21that's a pretty general statement. Do 2:23you do you want to cook? Do you want to 2:25do front of the house? You want to do 2:26back of the house? You want to get into 2:27investing and owning restaurants? Uh, do 2:30you want to just be a taste maker and 2:31hang out like Anthony Bourdain? Like, 2:33who knows? It's Anthony was a chef, but 2:35you get the idea. My point is that we 2:37need to get specific with career 2:39pathing. And that's what this video is 2:40designed to help get you to. First, 2:42let's get into some examples. An AI 2:44research scientist will need a PhD, will 2:46need published papers, and Meta is going 2:48to pay them a lot, right? Like Meta 2:49might pay them half a million dollars. 2:51An AI prompt engineer needs strong 2:53writing, doesn't need coding. They're 2:55not going to make as much as the meta 2:57seller. They might make around six 2:58figures, but they also don't need the 3:00same background. A machine learning 3:01engineer, wow, that's showing tremendous 3:04growth year-over-year, but it is a 3:06technical role, and you have to lean in 3:07on the technical chops. Meanwhile, AI 3:09Coach shows even stronger year-over-year 3:11growth at I think almost 60% according 3:14to the research I did, but it doesn't 3:16require the same technical chops as a 3:18machine learning engineer. These are 3:19entirely different roles with different 3:20prerequisites. So, I want to get at ROS 3:23specific qualification assessments that 3:25help get you an idea of where you stand 3:27and that's the project that I've been 3:28working on. So, let me set it up for 3:30you. I built 17 different assessment 3:34prompts. Each one is designed around a 3:36core insight about prediction. There's a 3:39piece that came out over the weekend 3:41from A16Z about how prediction is 3:44replacing postmodernism as the defining 3:46framework of our era. The idea is 3:49simple. Value creation is now about 3:51being predictive of the game rather than 3:54being predicted by the game. And that's 3:56exactly what these career assessments 3:58set out to do. They help you be 3:59predictive of your career path rather 4:02than being predicted by it. And that may 4:04sound like philosophy mumbo jumbo, but 4:07it's really relevant. I find the people 4:09who are able to make progress in AI are 4:12people who are able to accurately show 4:14intent and bridge that intent forward 4:17into the future by making specific bets 4:20on themselves. And what all I'm trying 4:22to do is set up prompts that help you 4:24make that kind of bet on yourself. The 4:25alternative is really simple. It's 4:27wasting 6 months, 12 months, or more 4:29during a really high leveraged period in 4:31human history pursuing goals that don't 4:33fit you well. And that's what happens 4:35when we just generically pursue AI. The 4:38old postmodern career advice was 4:40everyone gets a personalized version of 4:42their career, right? Take this boot 4:44camp, learn this framework, you get a 4:45unique resume, you'll be fine. That's 4:47not really true anymore. And we're 4:49seeing that breakdown. What matters now 4:51is timing relative to the AI revolution. 4:54where you engage and when you engage 4:56with the AI snowball that is rolling and 4:58tearing down the hill. Are you early on 5:00AI governance when the EU AI act makes 5:03it the hottest role next year or are you 5:05late to entry-level software engineering 5:07when those positions are down a bunch of 5:09percent. The assessment tells you where 5:11you stand and that's what I set out to 5:13do because then you can make a bet about 5:15where you want to go. So, how did I 5:16build these prompts? Building the 5:18prompts required me to think about 5:20prediction and anticipation deeply. I 5:23took each prompt and structured it out 5:25as an eight question interview, but the 5:27questions are not random. They're 5:28designed to extract signal about your 5:30qualification level from relatively 5:32limited information. Here's how I think 5:34about it. A good prompt anticipate what 5:37matters and it anticipates what doesn't. 5:39So for an ML engineer assessment, I 5:42don't ask, "Do you know AI?" I ask, 5:44"Rate your Python proficiency 1 to 10 5:47and describe your experience with 5:48TensorFlow or PyTorch. Have you deployed 5:50models to production?" That second part, 5:52the deployment piece, that's heavy 5:54signal. Lots of people will take a 5:56course. Relatively few will deploy to 5:58production with something meaningful. 5:59For an AI prompt engineer, I ask, can 6:01you share an example of heavily editing 6:04AI content to publication quality? 6:07Because that's a real skill. Anyone can 6:09generate the content. I've talked about 6:11this. The value is in the editing. The 6:14prompt anticipates the gap between what 6:17people think the job is and what it 6:18really requires. for AI governance 6:20specialists. Another one I ask about the 6:22EU AI act specifically, not do you know 6:25regulations in general, but the EU AI 6:27act is one of the biggest pieces of 6:30regulation on the planet around AI right 6:32now. Any company anywhere in the 6:34European Union, whether based there or 6:36not, has to deal with it. It's the 6:38hottest governance role in the new year. 6:41And the prompt anticipates the market 6:43reality. So these questions are 6:44structured to predict one of four 6:47outcomes. either you're qualified now 6:49for the role, you're almost qualified, 6:51there's significant gaps, or it will 6:53just say you're straight up not viable. 6:54And I think the honesty is helpful. And 6:56also each category gets to a timeline. 6:58And that's where the prediction piece is 6:59there and it gets a little fuzzy, right? 7:02Like 3 to 6 months for nearly qualified. 7:04Some people speedrun that with AI, 7:05right? 6 to 18 months for significant 7:08gaps. That's fair, but again, people can 7:10sometimes accelerate that if they lean 7:12into AI really hard. One of the things 7:14we forget is that AI is a kind of super 7:16skill. And if you learn how to learn 7:18with AI, you can speedrun some of these 7:20challenges. So the prompt doesn't just 7:22assess current state. It predicts the 7:24path forward based on patterns that I've 7:28seen in real career changes in the data 7:29in 2024 and 2025. 7:32It's not theory, it's this has worked 7:35for people with your background. Let's 7:36look at a pro. Okay, this is an AI 7:38product management prompt. As you can 7:40see, I have the same idea. You're 7:41setting up the interview at the top here 7:43and you just want to ask eight targeted 7:46questions you're looking to get signal 7:48you can predict against. In this case, 7:49we're targeting product management. So, 7:51we ask about overall PM experience. We 7:54ask about a IML concepts and we ask 7:57specific questions there, right? What is 7:59the difference between supervised and 8:00unsupervised learning as an example? I 8:03want to make sure that you're not just 8:05rating yourself 1 to 10 and kind of 8:06guessing or using cluyly to answer that 8:08you're actually able to be honest about 8:10it. We go through, we talk about 8:11stakeholder management, strategic and 8:13analytic skills. We talk about 8:15leadership and communication. One of the 8:17things I want you to notice AI product 8:19management is a great example of a role 8:20that's transforming and evolving. Right? 8:22We are looking at classic areas of 8:25product management expertise that are 8:28evolving and changing in the age of AI. 8:30And I think part of the way forward is 8:32to map them to specific competencies. 8:35And so we start to look and say how do 8:37we take classic experience with stuff 8:39like product strategy and road mapping 8:41and start to crosscalibrate it with how 8:44you've built stuff with AI capabilities 8:46in the product. And so a lot of the 8:48magic of this assessment is actually 8:51putting it together analyzing the 8:53responses and assessing how you qualify 8:55as an evolved role where you have both 8:58the PM piece and the AI piece. And then 9:01there's a personalized road map at the 9:02end because I want to leave you with 9:03something that's actually actionable, 9:05right? What are some recommendations 9:07that you can take with you? And this is 9:09useful by the way because 9:12you can depend on language models given 9:14this kind of prompt to go and search the 9:18web for the most relevant current 9:21courses and certification. I could have 9:23hardcoded all of these and I chose not 9:25to because the space is moving so fast. 9:28And so instead, given that we have 9:30reasoning models that search the web 9:32effectively with a very strong anchor 9:34prompt like this, when it comes time to 9:36recommend, they are going to look across 9:38your response set and be able to 9:39generate a custom list of courses for 9:42you that reflect your specific answers 9:46in a way that hard coding never could. 9:48And that's a reflection of where we're 9:49at with language models at this point in 9:512025. It's incredible. like you can get 9:53them to actually think through all of 9:54this and on the-fly recommend courses as 9:57long as you have a solid prompt. The 9:58salary reality check. This is one based 10:01on current market conditions. I could 10:03have also soft coded this and had them 10:05go the the prompt go out and benchmark 10:08the salary. I decided that was not worth 10:10it. I would rather prioritize the 10:12learning piece with the LLM tokens than 10:14the salary piece. And we can always 10:16update that over time, but it is 10:18reflective of where the market is at 10:20this point in the US. you'll need to 10:22discount it some for the EU. Finally, we 10:24get to the assessment criteria. I want 10:26you to understand how I assess this 10:27stuff so it's bold and transparent. I 10:29also want the LLM to be consistent about 10:31how to assess it. And that's it. Begin 10:33now by introducing yourself. So there 10:35that that's the prompt. It's actually 10:37not a magic prompt. I'm a big fan of 10:39showing how I prompt so you don't think 10:42I just do this stuff magically and it 10:44works. I want you to understand how it 10:45works. You know, the hardest part of 10:47building these prompts was building in 10:48anticipation of what people don't know. 10:51They don't know. Most people applying to 10:53AI roles don't realize how transferable 10:55some of their skills are. A compliance 10:57professional doesn't recognize the 10:59transference potential to AI governance 11:01necessarily, and if they do, they don't 11:03know which skills transfer and which 11:05don't. Someone with 5 years of change 11:07management might not realize they're 11:08qualified for an AI coaching role with 11:11tremendous year-over-year growth. So, I 11:13structured questions intended to surface 11:15those realities. I didn't just ask, "Do 11:17you have AI experience?" I asked about 11:20the background, the real background, and 11:22the prompt explains how it maps. You 11:24have compliance background. That's a 11:26scarce resource, right? You can get 11:27there faster than a software engineer 11:29could. This all connects back to that 11:31prediction framework. We create more 11:34order in the universe by contributing 11:36information. That's sort of the central 11:38thesis of the A16 Zpoint. It's been kind 11:40of bouncing around in my head. And I 11:42think that part of what we're doing with 11:43prompts like this is we are collecting 11:46information about what you actually have 11:48and bring to the table. Not what you 11:50think you're missing, but what you 11:51actually possess. And then we predict 11:53the most efficient path to contribute 11:55that information into the AI job market. 11:57That is a path to value. That is a path 11:59to career success. And that's why I 12:01built this. The assessment also 12:03anticipates mistakes. I designed it to 12:05call out common failures, analysis 12:07paralysis, targeting the wrong role, 12:09unrealistic timelines. There's a lot 12:11that potentially derails people and the 12:13prompt strives to get after failure 12:17modes and address them up front. Here's 12:18what I keep coming back to. The A16Z 12:21piece talks about how being good at the 12:22internet became a monetizable career 12:25because prediction is an the entire meta 12:28game. It's the entire game that we're 12:30playing in the economy right now. We 12:32literally share memes with probability 12:35prediction for game results, right? like 12:38I am a Seattle Mariners fan and there 12:40were memes circling about the 12:42probability of the Mariners winning this 12:45past weekend. And that's not the first 12:46time I've seen that. I've seen that for 12:47game after game after game, sports team 12:49after sports team after sports team. 12:51Prediction is a meme. The farther 12:53upstream you are in that information 12:55flow, the more likely you will be 12:57predictive of the game as opposed to 12:59being predicted by the game. And that's 13:00where these what these assessments do 13:02for AI careers. They put you upstream in 13:05the information flow. Instead of 13:06spending 6 months learning Python and 13:08then realizing ML engineer isn't viable 13:10with your background, you know it very 13:12quickly in like 15 minutes. Instead of 13:14applying to a 100 jobs and getting 13:16nowhere, you know more of which door to 13:18try first based on real market data and 13:21an AI customized conversation. So maybe 13:23you're going after the 41% growth in ML 13:26engineering roles or even the 134% 13:29growth in AI content creator roles. 13:30That's insane. You're being predictive 13:32instead of being predictive. You're 13:34contributing information about your 13:35actual qualifications to the market 13:37rather than getting lost in generic 13:38advice that doesn't account for who you 13:40are. And this matters more now because 13:43of that entry-level job displacement 13:44that I talked about at the top of this 13:46video. The old path was get your 13:48entry-level job, learn on the job, 13:50advance that is disappearing. The new 13:53path requires you to be strategic about 13:55which door you try first, and that 13:57requires prediction. But there haven't 13:58been very many resources about it, and 14:00that's why I created these prompts. I 14:02identified four overall pathways, if you 14:04want to ladder all of this up, that work 14:06in late 2025 headed into 2026. There's a 14:10technical bridge for people with a 14:11coding background into AI. It's a 14:13timeline of 6 to 18 months, depending on 14:15how fast you learn. There's a 14:17non-technical entry role for writers and 14:19creatives. That one's faster. It tends 14:21to be 1 to six months if you really lean 14:23in. There is a domain expert pivot, and 14:25that one really varies by your 14:27experience level with AI and the exact 14:29domain you're in. It's tough to predict, 14:31but from what I've seen, 3 to 12 months 14:33is pretty reasonable. And then there's a 14:35whole burgeoning exploding governance 14:37and compliance uh route. And that one is 14:39fairly linear. It's about 3 to nine 14:41months if you have some prior experience 14:43in any of the compliance roles. The key 14:46is those pathways have different timing. 14:48That's the prediction element. If you're 14:49a compliance professional, you can 14:51become an AI governance specialist a 14:52whole lot faster than you would become 14:54an ML engineer. And so the prompts help 14:57you predict the pathway that fits your 14:59background. and then give you a little 15:01bit of a sense of the timeline. And they 15:03might not give you what you want to 15:04hear, right? I wrote them to be honest, 15:07not to be comforting, because I think 15:09that's what we need in this moment. I 15:11wrote them to be true based on what we 15:13see in actual 2025 hiring requirements. 15:16The market is real. Amazon posted 15:19hundreds and hundreds of AI positions 15:22just in the first few quarters of 2025. 15:24Apple ditto, Tik Tok ditto. Everywhere 15:26you look, the AI related roles are 15:28exploding. But everyone's talking about 15:31AI like a singular thing. You need a 15:33personalized approach that's based on 15:35the real market data in 2025. Something 15:37that allows you to start to be 15:40predictive rather than the one being 15:42predicted. You know, every prompt I 15:44write and every prompt you write is 15:45fundamentally about prediction. Given 15:47this chain of information, predict the 15:49next good thing. That's what good LLMs 15:51do. But prompts anticipate what the 15:53model needs to predict. as well. They 15:54structure the information flow. They 15:56extract signal. They account for what's 15:58missing. That's what these career 16:00assessments do. Given this person's 16:02background, predict their qualification 16:04level. Predict their optimal path. And I 16:06had to anticipate what information 16:08mattered. As I was constructing the 16:09prompt, the prompt acted as a bridge to 16:11help you put your intent together and 16:14predict across the future where you 16:16should go from a role perspective. I had 16:18to account for what PE people typically 16:20miss, the transferable skills, the 16:21market timing, the entry-level 16:22displacement, and I had to package all 16:24of that into a prompt so that the LLM 16:27could predict the future in a way that 16:30was useful for you. Building good 16:31prompts is really about understanding 16:34what you are trying to predict, working 16:37backwards to figure out what information 16:39you need to predict it accurately, and 16:41then building that bridge. That is the 16:43craft of prompting, and I wish more 16:45people understood it. But TLDDR, I built 16:4817 different role prediction prompts so 16:50that you aren't asking about AI as a 16:52whole. You're asking about specific 16:54roles. So if you want to learn more, you 16:56can head over to the Substack. I have 16:57all the prompts there. If you wanted to 16:59just dig into that prompt and dig into 17:01how I think about prediction and how I 17:02think about getting into the job market, 17:04I hope this video has been helpful as