Published November 11, 2026 · Last updated November 11, 2026 · 10-minute read
TL;DR
- An AI engineer builds products with existing models: LLM APIs, retrieval (RAG), evals, tool calling and agents. Training models is a different job.
- A working software engineer already has most of the skills. The gap is a handful of new patterns, learned by shipping.
- Demand is real, and AI has not caused broad job losses in the data so far, though entry-level hiring is weaker in AI-exposed roles.
- Courses teach the vocabulary. You pick up judgment faster around people who ship: AI meetups, workshops, hackathons and builder houses.
- Bengaluru and San Francisco run AI builder events most weeks, and most aren’t featured on Luma’s discover page.
To become an AI engineer, build small products with existing models, one skill at a time (API calls, retrieval, evals, agents), and learn alongside people who already ship them. If you already write software, you don’t need a new degree. You need a few weeks of focused building, a public record of what you made, and regular evenings at AI meetups and hackathons where the working knowledge lives.
Most advice on how to learn AI starts with a course list. I’d start somewhere else. I recently told a friend on his notice period to stop letting interview rounds define him, come to Bengaluru or SF for two weeks, stay somewhere cheap, and go to events every evening. See what people are building. The referrals and the perspective follow. That advice is the spine of this post, because AI engineering is a craft that is still being worked out in public, mostly in rooms full of people trying things.
What does an AI engineer actually do?
An AI engineer builds software products with models someone else trained, using LLM APIs or open-weight models, plus retrieval, evals and agents. The clearest early definition came from swyx’s essay The Rise of the AI Engineer on Latent Space in June 2023. His core observation: tasks that once needed years and a research team now just require API docs and a spare afternoon in 2023.
That essay draws an API line. Researchers sit on one side, training and hosting models. AI engineers sit on the other side, turning those models into products. The essay also says AI engineers should treat product-specific data and evals as their own job, which is the part most tutorials skip. It quotes Andrej Karpathy: “One can be quite successful in this role without ever training anything.”
The supply argument is simple. The essay estimates ~5000 LLM researchers in the world, but ~50m software engineers. Companies can’t hire their way into research talent, so the work flows to engineers who can build around the models.
Here is what that looks like in a normal week: - Prompts and structured output: getting a model to return JSON your code can trust. - Retrieval (RAG): finding the right documents and fitting them into the context window. - Tool calling and agents: letting a model call your functions and chain steps together. - Evals: a test set that tells you whether the last prompt change made things better or worse. - Cost and latency: choosing the model that’s good enough, then caching everything you can.
If that list looks like backend engineering with a strange new dependency, you’ve read it correctly. (The strange dependency just happens to change its answer when you rephrase the question.)
Is AI engineering in demand in India and the US?
Yes, and a clear recent signal comes from job postings, especially in India. IANS, reporting LinkedIn data, said AI engineering job postings in India grew 59.5 per cent year on year, and named Bengaluru and Hyderabad as the main metros for AI hiring. A year earlier, LinkedIn’s Jobs on the Rise list for India put Artificial Intelligence Engineer third among the country’s fast-growing roles.
The tools have already arrived on most teams. In Stack Overflow’s survey of developers, 84% of respondents said they were using or planning to use AI tools in their development process. So “knows AI” is quickly becoming table stakes for any engineer. The people who stand out are the ones who can build an AI feature that holds up when real users touch it.
Search data tells the same story from the other side. Upskilling, learning AI and how to become an AI engineer are steady searches in both India and the US. The demand to learn is as loud as the demand to hire.
Will AI replace software engineers?
The current evidence says AI has not caused broad job losses, but it is squeezing entry-level hiring in exposed roles. Stanford Digital Economy Lab researchers, using ADP payroll data covering millions of U.S. workers through June 2026, state: “We find no evidence of widespread, economy-wide job displacement.” The same paper finds that employment of young workers (ages 22–25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with less-exposed peers, mostly through reduced hiring.
The productivity story is also less clean than the hype. In a METR study, 16 experienced developers working on their own open-source repositories took 19% longer to complete issues when allowed to use AI tools. They had expected a 24% speed-up, and even afterwards they still believed AI had sped them up by 20%. Trust is shaky too: Stack Overflow found more developers distrust the accuracy of AI tools (46%) than trust it (33%).
Read together, these point one way. Writing code is getting cheaper. Knowing what to build, and checking whether the AI output is right, is getting more valuable. An AI engineer is the person who does that checking for a living, which is why evals sit at the center of the job.
Key takeaways - AI engineers build on models; ML researchers train them. - Your existing software skills cover most of the job. The new parts are retrieval, evals, tool calling and agents. - Stanford found no economy-wide displacement, but young workers in AI-exposed jobs are falling behind. - Evals, meaning proof that the output is right, are the skill that separates a demo from a product.
How do you become an AI engineer if you already write software?
Pick one small product and add one AI skill to it each week, in public, while attending one AI event a week that matches what you’re learning. A course gives you the words. A shipped project gives you the questions. And an event gives you people who answered those questions last month.
Here is a plan for about two months, with roughly one skill and one kind of event per stretch.
| Weeks | Skill to build | What to ship | Event to pair it with |
|---|---|---|---|
| Week one | LLM API calls, structured JSON output | A small bot or CLI that does one useful thing | An AI meetup or demo night, to see what others build |
| Week two | Prompting and tool calling | Let the bot call your own functions | A hands-on workshop run by a model or tooling company |
| Weeks three and four | Retrieval (embeddings, chunking, search) | Q&A over a real set of documents | A builder-house evening, where people debug RAG out loud |
| Week five | Evals | A small test set of real inputs plus a score you track | A technical talk or paper-reading night on evaluation |
| Weeks six and seven | Agents | A multi-step task the model plans and completes | A weekend hackathon, as a stress test with a deadline |
| Week eight | Shipping and explaining | A deployed version and a short write-up | A demo day or show-and-tell, then ask for referrals |
Two rules make this work. First, ship something every week, even if it’s ugly, because a GitHub repo with a README shows what you can do in a way a certificate can’t. Second, write down the bugs. My retrieval returned the right document but the model ignored it is exactly the kind of sentence that starts a good conversation at a meetup.
Skip the math-first route unless you want to train models. You’ll still meet tokens, embeddings and probability, and you’ll learn them faster because a broken feature forces you to.
Why learn AI in the room instead of only from courses?
Because AI engineering changes faster than courses can be recorded, and current good practice lives in the heads of people who shipped something last week. The job title itself came out of meetups. swyx wrote that the AI Engineer Summit was announced after months of organizing small meetups, as a place for builders to compare notes.
A course tells you what RAG is. A meetup tells you that the team two seats over gave up on their vector database and went back to keyword search, and why. You can’t get that second sentence from a syllabus. And it’s usually the sentence that saves you a week.
This is the advice I gave my friend, applied to AI. Be in a good city and understand where the culture comes from. Two weeks of evenings in Bengaluru or SF shows you which problems people actually work on, which tools they’ve dropped, and who is hiring. Interview rounds test what you memorised. A room full of builders shows you what’s worth learning next.
There’s also the plain career logic. Referrals come from people who have seen you think, and a cold application rarely shows anyone that. A few evenings of asking sharp questions at AI events puts you in front of working engineers, and people tend to remember whoever asked the good question about eval sets.
Where are the AI builder events in Bengaluru and San Francisco?
In Bengaluru, follow AI House, hsrhackerhouse, AIBoomi, AI Garage, GPU Calendar, The Ecosystem Community, BLRxZo and AI Engineers Day (OpenAI and Peak XV); in San Francisco, follow AGI House, lablab.ai Events, South Park Commons and Inception Studio. Hidden Events tracks all of these calendars, so their public events appear in the city lists even when Luma’s discover page skips them.
Bengaluru’s top event category right now is AI, with 48 upcoming events tagged AI out of 127 upcoming public Luma events. Of those, 82.7% aren’t featured on Luma’s discover page. Most of the city’s events cluster in HSR Layout and Koramangala, so stay near one of them and your evenings get much easier. Start with Bengaluru tech events and Bengaluru meetups, and read our Bengaluru events guide for how the scene is spread out.
San Francisco has 72 upcoming AI events among 344 public Luma events, with 78.5% not featured on discover. The clusters are the Financial District, SoMa, Dogpatch and Potrero Hill. Use the SF AI events list and SF meetups, and check what’s on in SF this week.
| City | AI calendars to follow | Where events cluster |
|---|---|---|
| Bengaluru | AI House, hsrhackerhouse, AIBoomi, AI Garage, GPU Calendar, AI Engineers Day | HSR Layout, Koramangala |
| San Francisco | AGI House, lablab.ai Events, South Park Commons, Inception Studio | Financial District, SoMa, Dogpatch, Potrero Hill |
Not every AI event is worth an evening. Our guide on how to spot a high-signal tech meetup helps you skip the sponsor-heavy mixers, and how to find hackathons covers the weekend builds.
What should you do this week?
This week, pick one small AI project, ship a rough version, and RSVP to two AI events in your city where people demo what they’ve built. Bring the project on your phone. I built a bot that answers questions about our team’s docs, and retrieval keeps pulling the wrong page gets you a better conversation than I’m learning AI.
Bengaluru’s event calendar peaks on Saturday and SF’s on Tuesday, so plan your week around those. If you’re between jobs or serving a notice period, do what I suggested to my friend: two weeks in the city, a cheap room, and an event every evening. Start from the Bengaluru startup events or SF startup events list and go from there.
About the author: Himanshu Saleria, founder of Hidden Events. I love meeting people. I’ve been attending events, hosting them, and picking up patterns along the way for what makes a gathering worth showing up to. Hidden Events surfaces what I’ve been finding. LinkedIn
How we research: event counts come from Hidden Events’ own crawl of public Luma events, refreshed every 4 hours. Other figures link to their original source: Latent Space, The Rise of the AI Engineer, IANS via Telangana Today on LinkedIn’s AI hiring data, People Matters on LinkedIn Jobs on the Rise India, Stack Overflow Developer Survey, AI section, Stanford Digital Economy Lab, Canaries in the Coal Mine?, METR’s study of experienced open-source developers. Search volumes are from Google Keyword Planner.