you don't have to compare whether stochastic fuzzing or agentic testing is the best. You can just do both of them. They both have enough ROI to be worth the trouble. That's why I said "in addition to".
It’s 2026, so another basic technique that every team should add to their testing strategy (in addition to proven techniques like fuzzing) is agentic user simulation. Set up an AI agent with access to the product, and prompt it to use the thing in hundreds of realistic use cases, and to report any possible bugs. It will catch a lot of ‘blind spot’ bugs that were previously things that were only caught by humans.
Yeah I think this is a good approach. I’m pretty AI-optimistic when it comes to making code changes. But reading AI generated descriptions (including pull requests) is absolutely the worst. That content really needs to be human written. Not just for the benefit of the reader, but it also helps the writer exercise their understanding.
Slowing down development is the wrong goal. I see a desire for slowness come up a lot with developers. If you pursue that goal all the way to its logical conclusion then eventually you would stop all coding completely. Which would prevent new bugs but obviously we can't do that and keep our jobs.
By all means add tons of quality gates to your SDLC pipeline. But thinking about slowness purely for the sake of slowness will not solve your problems.
- Skills do take up space in the context. The name and description of every skill goes into the system prompt. You can't add unlimited skills without context pressure.
- MCP context spam is less of a big deal now that there is deferred tool loading.
- If you work with the agents long enough you'll run into situations where using bash CLI tools suck and an MCP works better.
- MCP is not dead and has never been dead, it's the right solution in certain cases.
Good question but people are doing it anyway. It's a fact that right now tons of people are buying Mac Minis specifically for this use case, to treat them as their personal data center for agents. The concept of "power use for inference" is foreign. Those people are the ones that motivated this blog post I think.
That's a lot of assumptions. I think there are also people buying new hardware specifically for this purpose, and their motivation to do it is thinking it will be cheaper in the long run. Privacy is not necessarily the motivation.
The time difference does matter though. There were some recent worm attacks in NPM that spread very quickly because they used post-install. I don’t remember how long it took NPM to block the packages but it was probably around 30 minutes or so? If it wasn’t for post-install then that same attack would have a much slower spread and thus a smaller blast radius.
1) the minimum number of employees it takes to maintain the core product
Vs 2) All the employees that it makes sense to hire for revenue and market expansion.
Internet comments usually assume that (1) is the goal. But think of say the sales department. If every salesperson you hire brings in new company revenue that’s greater than their salary + overhead, then why not hire 1000 of them?
Offshoring has been a common practice for decades, it works great for some functions and not great for others. Why would it suddenly have a massive uptick in 2027?
In a world of AI coding I think rule 5 is still as important than ever. I don’t validate everything Claude does but I do put attention on data structure design since it’s so important.
Lol anyway, the point is that even in a scenario where all the major models disappeared tomorrow (including OpenAI, Anthropic, etc), we would still keep using the existing open source models (GLM, Deepseek, Qwen) for a long long time.
There's no scenario where AI goes away completely.
I don't think the "major AI services go away completely" scenario is realistic at all when you look at those companies' revenue and customer demand, but that's a different debate I guess.
> it's the running costs of these major AI services that are also astronomical
There's wildly different reports about whether the cost of just inference (not the training) is expensive or not...
Sam Altman has said “We’re profitable on inference. If we didn’t pay for training, we’d be a very profitable company.”
But a lot of folks are convinced that inference prices are currently being propped up by burning through investor capital?
I think if we look at open source model hosting then it's pretty convincing - Look at say https://openrouter.ai/z-ai/glm-4.7 . There's about 10 different random API providers that are competing on price and they'll serve GLM 4.7 tokens at around $1.50 - $2.50 per output Mtokens. (which by the way is a tenth of the cost of Opus 4.5)
I seriously doubt that all these random services that no one has ever heard of are also being propped up by investor capital. It seems more likely that $1.50 - $2.50 is the "near cost" price.
If that's the actual cost, and considering that the open source models like GLM are still pretty useful when used correctly, then it's pretty clear that AI is here to stay.
I've never worked at a company as large as Google but in my experience things can be a little more optimistic than the post. When earn enough trust with your leadership, such as at the staff/architect level, you'll be able to tell them they are wrong more often and they'll listen. It doesn't have to be a "$50,000 check" every time.
That leads to a very important question - Why doesn't leadership always trust their engineers? And there's a very important answer that isn't mentioned in the blog post - Sometimes the engineers are wrong.
Engineers are extremely good at finding flaws. But not so good at understanding the business perspective. Depending on the greater context there are times where it does make sense to move forward with a flawed idea.
So next time you hear an idea that sounds stupid, take a beat to understand more where the idea is coming from. If you get better at discerning the difference between ideas that are actually fine (despite their flaws), versus ideas that need to die, then you'll earn more trust with your org.
to some degree this is a "market correction" on the inherent value of these papers. There's way too many low-value papers that are being published purely for career advancement and CV padding reasons. Hard to get peer reviewers to care about those.
Yeah I don't know about that, the model providers like OpenAI, Anthropic, etc, literally sell intelligence as a product. And their business model is looking a lot more stable in the long term than all the startups built on top.