I'm Aniket, a software architect from Pune.
I spend my days designing systems, and I keep noticing the same thing: most tech problems are actually business problems in disguise.
Here I write about:
- How good architecture makes change cheap
- The trade-offs behind real product decisions
- The business mechanics of modern software
Simple words, deep ideas. If that sounds useful, follow along.
Why does a server at 90% busy feel so much slower than one at 70%?
Because waiting does not grow in a straight line. In the simplest queue model, the time a request waits in line is busy / (1 - busy), times the time the job itself takes.
At 50% busy, it waits about as long as the work.
At 80%, 4 times longer.
At 90%, 9 times.
At 95%, 19 times.
So the last bit of "efficiency" is exactly where the delays blow up.
This is not only about servers. A team planned at 100% capacity works the same way. Every urgent bug and small ask joins a line that never clears, and the business sees it as "engineering is slow".
Headroom looks like waste on a spreadsheet. In practice it is what keeps a system, or a team, fast.
How much slack does your team actually plan for?
@RahulbeGupta C, with one change. Most developers have the requirement. What they lack is the reason behind it. Ask why a feature matters to the business and you often find a smaller version that solves the same problem in a week, not a quarter.
@benmathew Parking has one goal and fixed rules. A pricing page is really a decision about who your customer is. Until that's settled, no layout will feel right.
@eddyvustg I'd add a third: narrow permissions per task. Tests and schemas catch a wrong shape. They don't catch a correct call made against the wrong data. An agent with read-only access for a research task can be wrong without hurting anyone.
@AKirtesh System design, because it makes the other two useful. DSA makes one piece fast. Agents now write the pieces for you. Someone still has to decide where each piece goes and what happens when one of them fails.
@kutluhaneth At 2am the missing piece is rarely the code. It's the reason. A two-line note on why something was built this way saves more time than any clever fix, and agents are never asked to write that part.
Finding bugs in a diff was never the slow part of review. The slow part is asking whether this change should exist at all, and whether it fits how the rest of the system works.
A tool reading one PR cannot see that the same logic already lives in another service. Free tools will make the first part cheap. The second still needs someone who knows the codebase.
The webmaster comparison is useful because of what happened next. The job split. Some people went into frontend, some into ops, some into SEO and product. The title vanished, but the work under it grew.
Same with DBAs. Managed databases took away backups and patching. Schema design, query plans and data modelling did not go anywhere. They moved to the backend engineer, who often learnt them the hard way during an outage.
So my guess is that typing code stops being a job, and the judgement around it spreads across fewer people with wider roles. The hard part is how juniors build that judgement if they rarely write code by hand. How would you solve that?
Show the outcome, not the risk. "This will refund βΉ4,200 to 37 customers" gets read. "This action may have side effects" gets clicked past within a week.
I'd also keep that preview in a log. After an incident, you want to see exactly what the human agreed to, not what the agent did later.
@OscarHartmann I'd ask which cities, not whether e-commerce recovers. A dark store is a fixed cost waiting for enough orders per hour. Dense Indian metros with cheaper delivery labour can fill that hour. Many European cities never could, whatever the funding mood. Did your 12 split that way?
@eliadele Agencies sell hours. So a tool that saves one hour turns straight into billable money, and the owner can do that maths in his head.
A developer who saves an hour just gets a free evening. Nice, but nobody pays $19 a month for a free evening
@Overlap_Tech And it's the part no tool speeds up by itself, because it's a negotiation between people with different incentives. Sales wants the feature for one big client, engineering wants the design that survives the next five. Cheap to type, expensive to get wrong.
@SteveJG03 The hours are the honest part. 1000 hours is close to half a year of one full-time engineer. And the bill doesn't stop at launch: every field and workflow is now theirs to maintain. Build vs buy is rarely decided in month one. It shows up in year three.
Part of it is where you sit in the value chain. An entrepreneur gets paid when the customer's problem goes away. An employee mostly gets paid for the skill of building the thing. AI makes building cheaper, so one person gets a discount and the other feels a price cut.
Same tool, two very different balance sheets.
The interesting middle is engineers who already own a number, like checkout conversion or on-call load. In my experience they react a lot like founders. So maybe the split is less about entrepreneurs and more about who is measured on results. Have you seen that in your conversations?
UPI is one of the most open payment systems ever built. Any app can pay any QR code. Your money stays in your bank, not in the app. Switching apps takes a minute.
And still, two apps handle nearly 80% of UPI payments.
This is a useful lesson for anyone who designs platforms. Open rails remove lock-in at the data layer. They don't remove habit. The lock-in simply moves up, to whatever people touch every day: the app on the home screen, the QR code at the shop counter, the speaker that announces every payment.
The money follows the same path. Most UPI payments earn the apps almost nothing, so they earn on what sits on top: loans, insurance, bill payments, ads. The more users you already have, the better that works.
The deadline for NPCI's 30% market-share cap is 31 December. But enforcing it means telling the biggest apps to stop adding new users. That is a rule trying to fight a habit.
If you were designing UPI again, what would you change to keep it competitive?
@elaikajoyce Cost of the status quo is the right anchor. Often the real competitor is an Excel sheet and one tired accountant, which looks free until you count the hours. Spending a day watching the customer work usually changes the price you're willing to ask.
@viralbshah And carrying risk needs a different kind of company. With hourly billing, a delayed project still earned money. With outcome pricing, the vendor needs people who can push back on scope creep and estimate honestly. That's a culture change more than a pricing change