Software engineering is becoming actual engineering: you design and review, AI does the construction. Tech layoffs are a ZIRP hangover, not an AI apocalypse. Only 6% were primarily AI-driven. The uncomfortable part: a lot of devs were doing fake work and AI is about to expose that. What survives is taste, knowing what to build and why. The good news: anyone with a good idea can now build it, and if it's wrong, scrap it and try again tomorrow.
That's not real engineering mate. That's God Complex rationalizing it's own continuing to do what it's always done while everyone else drops like flies. You'll know Software Engineering has started to turn into real Engineering when programmers/Sys Architects stop being unlicensed roles, and we start getting PE's without whose assent work cannot be started. It'll be real engineering once we've started cutting society so deep, they realize it is not enough to just let anyone build what they want and offer it as load bearing infrastructure.
Once we see that, then we can talk about being Big E Engineering. Until then, it's just people twiddling bits for ethically dubious reasons pretending they aren't violating regulations because they're using computers to do it.
If you can define what correct looks like in the form of failing tests, AI agents can do the work of making them pass. TDD always had the right instinct, but now there's cheap, tireless labor to actually exploit it. The hard part is that writing good tests requires deep domain knowledge, and AI isn't great at that yet. It'll generate hundreds of shallow unit tests that cover nothing meaningful. So the real skill shifts from writing code to writing specs. The implementation is the easy part now.
The thing protecting your software's secrets wasn't encryption. Tt was that almost nobody could read compiled code. Now AI can. For dollars. In minutes.
The few thousand people worldwide who could reverse engineer binaries were the entire moat protecting IP, firmware security, and software as a discipline. That scarcity is gone.
few thousand? this binaries being a protection has been changing for a long time already by drm, sgx and other means. there has been a steady increase in complexity and effectiveness in hiding stuff through drm/encryption and other schemes.
just that some LLM calls it a moat or rare skill is nonsense. there are whole industries in which its normal dayjob for people to reverse engineer binaries let alone for the thousands upon thousands that graduate technical programs each year and can do proper RE. then ofc many thousands of hobby people who are likely some of the most sharp RE people out there.. huge communities reversing games, consoles, firmwares....
I would say I'm not AI but that's what AI would say, right? There are 3 R's in strawberry!
There are several products on the market that are automating reverse engineering. Think of what people do: they orchestrate tools, build tools, read code, run things in emulators / sandboxes. All of that can be done with an agent in the middle. Building the tools is hard, knowing how to use them is kind of hard.
If the code isn't obfuscated / packed / etc, claude code and a ghidra mcp server is better than most of the reverse engineers out there and way, way faster. You can work to get other tools in the mix for claude code.
Are models getting dumber or smarter? Is code getting easier or harder to write? The writing is on the wall.
im not sure really. i do not think claude with an MCP is better than reverse engineers. I think that is taking some bad baseline of what reverse engineers are capable of.
Claude instructed by a highly skilled engineer might outperform a poor engineer, this is simple to see.
The innovation would be that claude guided by a shit engineer (or just a sales guy or my mom..) would outperfor a senior engineer. Senior in skills and experience, not title.
and RE is a vast field with an extreme wide skill tree. many of which skills are undocumented and not publicly shared,.using even tooling not available to these models...
its not just some dudes with ghidra lookin at firefox
.. thats the rarer case :S
Python was built around human limitations. AI agents don't have those limitations, so Python's trade-offs (loose typing, runtime errors, no compiler) are now liabilities. Rust's strict compiler, which humans hate, doesn't bother agents at all. Pick your language based on who's actually writing the code.
Negative temperature coefficients of reactivity matter to keep an operating reactor stable and prevent rapid overpower accidents a'la Chernobyl. All modern reactors exhibit this characteristic. However, the vast majority of risk in reactors today is that the fission energy doesn't all come out at the moment of fission. Some of it (roughly 7%) comes out after shutdown as exponentially-decaying radiation of the fission products. This decay heat (as it's called) is the primary safety hazard of today's reactors because there's enough of it to breach the reactor vessel if something goes wrong (a'la Fukushima). The Fukushima chain reaction shut down perfectly after the earthquake. The decay heat removal systems failed after the tsunami came along. The decay heat melted the fuel and cladding without cooling.
MSRs and other advanced reactors have passive decay heat removal by making use of different coolants (molten salt, liquid metal, etc.) and natural circulation air heat exchangers.
It has modules which look for patterns in code. Then you can tell it to run some method from the original app to understand what the code should be. Then, you can replace the obfuscated code with whatever you computed.
maybe you could make a chrome extension which blocks all forms of fun and frivolity. you could just hide images from various meme sites, maybe do some checksum comparisons?
on one hand, I completely agree with you. on the other hand, i've had to make presentations and talks and the stuff just looks so dry and boring and people respond surprisingly well to old jokes. if you don't include any sort of lulz they just sit there thinking they're learning, but if you mix in a few memes, it helps wake them up and they start to pay attention.
I think a p-hash[1] would work better at detecting the same image... but what about a neural net that detects fun? That would probably be pretty enjoyable to write.
