> No more AI thought pieces until you tell us what you build!
We build a personal finance tool (referenced in the article). It's a web/mobile/backend stack (mostly React and Python). That said, I think a lot of the principles are generalizable.
> Writing _is_ the thinking. It's a critical input in developing good taste.
Agree, but I'll add that _good_ prompt-writing actually requires a lot of thought (which is what makes it so easy to write bad prompts, which are much more likely to produce slop).
I know many companies that have replaced Customer Support agents with LLM-based agents. Replacing support with AI isn't new, but what is new is that the LLM-based ones have higher CSAT (customer satisfaction) rates than the humans they are now replacing (ie, it's not just cost anymore... It's cost and quality).
Well I as a Customer who had to deal with AI bots as Customer Service have significantly lower Customer Satisfaction. Because I don't wanna deal with some clanker. Who doesn't realy understand what I am talking about.
I call it L-ai-ziness and I try to reduce it on my team.
If it has your name on it, you're accountable for it to your peers. If it has our name on it as a company, we're accountable for it to our users. AI doesn't change that.
This sandboxes your file system. That's just one class of problem. People will want to hook this up to their inbox, their calendar, their chats, their source code, their finances, etc. File system secured? Great. Everything else? Not so much.
Getting the sharing in-place, yes, but maintaining it operationally would still be a headache. Things like schema migrations across shards, resharding, and even observability.
You are both legends. Your original MapReduce paper is what inspired me to work for Google (2006-2009), narrowly dodging a career as a quant on Wall Street.
This is something I've thought about a lot, and while I like the framing in the article, it's missing a few key dimensions.
Optionality: In addition to "letting things resolve themselves", one benefit you can sometimes get by deferring a decision (esp a "one-way door" decision) is optionality (of learning information that might result in a better decision).
Waffling: On the other hand, if you are a manager or decision-maker on whom others depend, one of the worst things you can do is waffle on a key decision (ie, be indecisive). Andy Grove has a paragraph on this in High Output Management as one of the highest negative leverage things a manager can do to their team, and in fact, often a wrong (but correctible) decision is far better than no decision.
Good managers instinctively know how to navigate these tradeoffs.
You can defer the decision without waffling on the communication. Saying "we'll decide this later" (ideally with some reasons / parameters), is decisive - and sometimes gets pushback. Going silent in order to avoid dissent is bad form.
Not to say this "technique" isn't useful, but imo it should be super limited. I'll put off a reply when a) it's a non-urgent issue, where b) resources (documentation, or other local users' experience) exist, c) this user will be motivated to find them (by the nature of the issue, or because I know they are the sort of person who does that), and d) independent problem-solving will fulfill a teaching function. Even then, I will hit "Snooze" on the email and follow up a day or two later: if they haven't solved it I'll point them to a resource; if they have I'll praise them for figuring it out for themselves. People like both outcomes.
I notice that many of the historical examples are a result of latency in communication, like people asking for things that had already been done. We don't often face that constraint.
Have you tried Starred Inbox? That turned me from "person who often slips on requests" to "person who rarely ever slips on requests".
Basically, keep your inbox as zero unread. If something should be dealt with it immediately, deal with it immediately. If it needs to wait, Star it. Now, your inbox has a list of all Starred messages at the top. At the start and end of work day, work through Starred items to either tackle them, or keep them Starred for later. But whenever you open your inbox, the starred items are at the top.
With inbox you could have multiple bundles, instead of one "starred" section, and each bundle could have a policy so that after you finished triaging it would automatically be hidden for a day or a week, so that it wouldn't constantly distract you by appearing at the top every time you glance at your inbox. This feature was also never ported to Gmail.
This was eye-opening. I used to think militaries were completely centralized and top-down, but a friend who was an officer explained this to me and pointed me to the literature. It was fascinating and educating to understand the principles behind Mission Command being successful as a method (competence, mutual trust, shared understanding, etc).
> Commanders, staffs, and subordinates ensure their decisions and actions comply with applicable U.S., international, and, in some cases, host-nation laws and regulations. Commanders at all levels ensure their Soldiers operate in accordance with the Army Ethic, the law of war, and the rules of engagement. (See FM 27-10 for a discussion of the law of war.)
It isn't very complicated from a military law perspective. The chain of command (following orders) has a lot more weight on it than a given solder's interpretation of military, constitutional, or international law.
If you believe you are being a given an order that is illegal and refuse, you are essentially putting your head on the chopping block and hoping that a superior officer (who outranks the one giving you the order) later agrees with you. Recent events have involved the commander in chief issuing the orders directly, which means the 'appealing to a higher authority' exit is closed and barred shut for a solider refusing to follow orders.
That doesn't mean a soldier isn't morally obligated to refuse an unlawful / immoral order, just that they will also have to pay a price for keeping their conscience (maybe a future president will give them a pardon?). The inverse is also true, soliders who knowingly follow certain orders (war crimes) are likely to be punished if their side loses, they are captured, or the future decides their actions were indefensible.
A punishment for ignoring a command like "execute those POWs!" has a good chance of being overruled, but may not be. However an order to invade Canada from the President, even if there will be civilian casualties, must be followed. If the President's bosses (Congress/Judiciary) disagree with that order they have recourse.
Unfortunately the general trend which continues is for Congress to delegate their war making powers to the President without review, and for the Supreme Court to give extraordinary legal leeway when it comes to the legality of Presidential actions.
The YPG, an armed anarchist military group, defeated Islamic State in North East Syria and more or less founded their own country without a traditional military command structure. Instead they had loosely coordinated teams.
We build a personal finance tool (referenced in the article). It's a web/mobile/backend stack (mostly React and Python). That said, I think a lot of the principles are generalizable.
> Writing _is_ the thinking. It's a critical input in developing good taste.
Agree, but I'll add that _good_ prompt-writing actually requires a lot of thought (which is what makes it so easy to write bad prompts, which are much more likely to produce slop).
reply