Work on robotics or agents?

Pick the immediate direction that will put you in a position to work with the smartest people. That’s probably more important than picking a technology.

Your side projects with agents have impressed people in your org. Do those people impress you? My gut reaction is that if they are impressed by the side projects of a recent college grad, they may not be at the top of that field.

At an early stage of your career, the best work environment is one that makes you feel like “damn I’m really going to have to perform to keep up here.” It’s not great to be very early in your career and feel like “damn, I’m the smartest guy in this room.” It can create bad habits and a sense of entitlement.

And the tricky thing is, if you are a high performer, most situations will make you feel like the latter.

Different take here -

In my career it hasn't mattered if I'm cleaning bathrooms or building Software used by millions of people IF I'm doing it with people I enjoy being around. The people you work with will make or break your career, and (this will be shocking) if you're smart and like working on hard problems you'll end up wanting to be around people who are like that... so choose your direction based on the people.

If space lives up to even a fraction of its potential over the next ~10-20 years then robotics is going to grow astronomically as a field. More generally I have not found software to be a longterm satisfying career - the compensation, freedom, and potential are all excellent. You are well compensated, can work from anywhere*, and a single successful side project (don't sign away your rights to code you write outside of work) always has the possibility to blow up into something worth billions - this is really unlike any other "normal" career. But I think there's a reason that there's a growing sect of people leaving to do everything from farming, woodwork, and even welding. Building even simple things in the real world is somehow so much more satisfying.

If I had it to do over again I'd pick a field where software complements something done in the real world - robotics, astronomy, aerospace, etc.

Successful software career turned robotics (2015) here. After a crypto exit (unicorn) I committed 100% to robotics, driven by sheer boredom with software, nail-in-coffin typified by the perception of the crypto world's journey from utopian vision to sad elucidation of status quo slash enshrined large capital slash regulator interest. Also opportunity: had already learned Chinese, knew the domestic business environment, had a global outlook and felt this was a unique starting position.

The switch has been a very different run to most software people heading to robotics, because I did this without external capital in China where it is cheap and small amounts of money (relatively) are unreasonably effective due to higher iteration speeds and lower supply chain latencies and costs. Having thus learned a heap running industrial facility, including equipment selection, maintenance and integration plus managing cross-disciplinary teams across all aspects of mechanical, electronic, electrical and production/process engineering, we are about to raise for US-based go to market, with a target of outright purchase of a permanent R&D slash autonomous manufacturing facility in San Diego so we can punch holes in walls and get core process really humming.

If you are interested in multi-disciplinary design work spanning software, hardware and operations research (think "real life Factorio") please reach out. We have some interesting problems and seek to buid a US-based core of talented generalists with what we would hope is a PARC-style pragmatic engineering culture (whole problem in view) rather than corporate-style "fill in the blank" grind. Positive environmental outcomes such as avoidance of single use plastics are part of our values. Prospective early 2Q start. Email in profile.

Robotics hands down. Robots exist today with demonstrated value and are on every bit the hyper-growth curve in reality that LLM wrappers claim. Robots build everything from printed circuit boards to automobiles to other robots.

The term “agent” isn’t really even defined out of some effort to fundraise by establishing some price target anchor to NVIDIA.

The really honest and serious people in a space that could be called agents are people like Nick and Adrian at Cohere, and historically they’ve been both specific and honest about what they’re doing, they’ve often called it “tool use”, which is a real thing. If you want to work on agent type stuff I’d go talk to Cohere.

Nadella saying on camera that “AGENT” will replace all business software near term is all you need to know: Nadella is smart as hell and wildly well-informed. If he says outlandish shit like that?

He’s lying.

Pick the team that most impresses you with their expertise. That sounds like the robotics company.

Are you in it for the glory, for the money, for the experience, for the team and colleagues, for the resume, for the challenge, or for something else? Answer that question and you have your answer.

Why not both. Isnt that where it is headed anyway? Agents are hitting the S-curve in terms of raw text so the next big step is robotic agents. Nvidia is clearly pointing the way here.

In my opinion these fields will merge at some point in time as nextgen LLMs can be used to reason trajectories.

Consider RoboAgent[1] or LBMs at TRI[2]

[1] https://robopen.github.io/media/roboagent.pdf

[2] https://pressroom.toyota.com/toyota-research-institute-unvei...

I think from a technological perspective, building actual hardware has more relevant tangible real-world impact. But in the end, as many other posters have said, it really depends on the team, their experience and their capability to build product/drive business.

Shameless plug, I have been writing about this since 2023. [3] https://jdsemrau.substack.com/t/robotics

Robotics has less people involved in and less hype at least in open source GitHub projects. The good part that there is still lot of low hanging fruit to harvest even without touching real hardware. Namely simulators, models (most don't need multimillion dollar rigs) and infrastructure / robotic os projects such as ros.

Agents, LLMs are real cool stuff these days but it is too early to say that this topic won’t fade in next few years if we can’t return to ZIRP.

Robotics on the other hand is entrenched segment and will always be there and will probably gain popularity covertly as automation is de-facto goal of all businesses.

Lastly, keep the Agents as side project(org people getting impressed by agent workflow tells more about them than you) as when/if Agent hype collapses, you are sure to be gutted, but robotics is probably the core business and long term secure. Eventually we will need more robotics(production automation, natural disaster recovery, rising global conflicts, medical surgeries etc.) but we may get these agents replaced by some other ground breaking power efficient models in next few years.

You’re a mech engineer so your passion is most likely robotics. You’re young and can take risks and possibly, financial risks.

Contrary to what the cynics here say, LLM agents are in fact a thing. They are already reducing hiring and will only further disrupt the labour market for white colar jobs.

I propose an middle ground. Forget _mechanical_ engineering. There’s no money in robots unless you’re building roombas or industrial scale assembly line machines.

Aim to be the person who gives a robot _agency_ something like https://deepmind.google/discover/blog/shaping-the-future-of-...

Without additional qualification, the word "robotics" is meaningless. I've worked in automation of physical processes for my entire career. Some people refer to stuff I've done as robotics, I don't.

Both security and robotics will continue to progress quickly (I'm actually taking a break from writing some cybersecurity documentation rn) because they address real, pressing problems.

So do the one that sounds more interesting. They'll both likely be very lucrative.

I’ve been in robotics 12+ years now and I feel that having something tangible that my software works on is incredibly important for keeping my job interesting.

Robotics is an unforgiving field, and is generally not a sustainable commercial business outside research. However, if you live in a physical factory focused economy that might differ, but generally in north America it is niche field.

Security is also a niche field, but at least it is a more common lucrative position.

"An expert knows more and more about less and less until he or she knows everything about nothing."

=3

Whatever the one you want to do is, but you're not sure because of the other one...ignore the other one, do the one you want to do.

I used to work on robotics, then left to focus on AI - now without question robotics is the place to be.

I’d go for robotics. The whole agent thing while not entirely without merit seems quite heavy on hype without necessarily add that much more over the underlying LLM magic.

What is a "security-related agent"? A robot guard?

Pick the area that has more engineering complexity and complements your unique skills and personality. In this case that would be robotics

you should do robotic agents :-)

Seriously though, i think working at a startup at an early stage in your career is a great opportunity to work on a lot of things with lots of dedicated people.

On a side note, robotics is an exciting field as i do think robots gonna be everywhere in the not so distant future, from advanced drones, to agricultural robots, to humanoid robots, etc...

agents workflows etc. sounds like something very perishable, easily replaceable, shallow

Robotics is "hard".

Other things being equal - hard things tend to make much more impact on your professional growth and look more impressive in CV.

Work wherever you like and respect the people more.

If you end up going with robots in SF look me up!

Robotics, without a doubt. Those exist today and are an actually feasible technology. Agents will very likely die out in the not-too-distant future, once the hype dies down and some more lawsuits go around and companies realize that "99% accuracy in production" is just pathetically low for these kinds of applications. At scale that's literally thousands to millions of things going wrong per second.

Offshoring and AI …. Twin threats to all IT jobs including SWEs.