Afsar Ahmad

AI models have benchmarks. Human+AI teams mostly don’t--even though that is how more professional work is getting done. Most professional work is becoming human+AI. Yet many interviews either prohibit AI or test knowledge that candidates can retrieve instantly while working. I’m exploring a different assessment: • Give a person a realistic, unseen task. • Let them use the same AI, tools, time and compute available to everyone. • Run AI alone on comparable tasks as a versioned baseline. • Evaluate the final quality, error detection, judgment, verification, consistency, time and cost. The question is not, "Who writes better prompts?" It is: How much useful capability does this person add beyond what the AI can produce alone? I don’t yet know whether this would predict workplace performance better than an ordinary AI-enabled work sample. That is the premise I want to test. The first application may not even be hiring. It could be helping companies understand: • Which teams work effectively with AI? • Who catches AI failures? • Where is training needed? • Does #AI training actually improve performance? What is fundamentally wrong with this idea? What evidence would make you trust--or reject--such a benchmark? #FutureOfWork #Hiring #HRTech #HumanAICollaboration

8

Tired of finding performance issues in production? ApexGuru is now available as an MCP server inside your Agentic IDE. 👉 Run Apex scans without leaving your editor 👉 Get insights backed by actual runtime data, not just linting rules 👉 See what's likely to blow up before you push Shift-left, but for real this time. 📖 Blog: https://lnkd.in/gMKV-rrR Big shoutout to Mayuresh Verma who co-authored this with me. If you're curious about the reasoning behind the design decisions, it's all in the blog. Note: Check out the Demo in the comments! 👇 Curious to hear how your teams are thinking about "Performance-First" development.

31

1 Comment

UPI MDR. My first reaction after hearing about this was simple: This is bullshit. Not because I don’t understand that UPI infrastructure has costs. But because I don’t like the direction this is taking. Let’s break it down. Under the new framework, a 0.4% MDR will apply to specified P2M UPI transactions above ₹2,000, subject to a maximum of ₹300 per transaction. Importantly, this is a merchant-side charge, not a direct fee charged to customers. So technically: Customer → ₹5,000 UPI payment → ₹5,000 debited No additional ₹20 “UPI fee” appears for the customer. Sounds fine, right? I’m not convinced. Because businesses don’t magically absorb additional costs. They either: → absorb it and take a margin hit → reduce discounts → increase prices → or find ways to pass the cost down the chain And eventually, the customer can end up paying for it indirectly. That’s my biggest concern. UPI became incredibly powerful because it made digital payments almost frictionless. Scan. Pay. Done. No thinking about transaction fees. No worrying about whether the shop accepts your card. No cash. And now we’re introducing a cost into one of the world’s largest digital payment ecosystems. I understand the argument that UPI needs to become economically sustainable. Banks, payment apps, PSPs, fraud prevention, cybersecurity and infrastructure all cost money. Fair. But my question is: Why does every successful digital ecosystem eventually need to be monetised? Today, we’re told: “Don’t worry. The customer isn’t being charged.” Maybe not directly. But if the merchant’s cost goes up, who do you think ultimately absorbs that cost? That’s where I disagree with the direction. I don’t want UPI to slowly become another payment system where small charges become normal and eventually get passed on to consumers. Because once we normalise the cost, it becomes very difficult to go back. UPI was one of India’s biggest digital success stories because it removed friction from payments. Let’s be careful that we don’t slowly put that friction back. What do you think, will merchants absorb the MDR, or will customers ultimately pay for it through higher prices? #UPI #MDR #DigitalPayments #Fintech #India #NPCI #DigitalIndia #Payments

5

Your Apex tests aren't catching bugs - this plugin can prove it. Over the past year Sébastien Colladon and I have been chipping away at an #opensource Mutation Testing tool for Apex. It introduces small bugs into your code ("mutants"), re-runs your tests, and reports whether your test caught the bug. More on how it works below. The idea behind mutation testing isn't new, it's been used in other languages for decades. But we finally have a tool to bring it into Apex, so you can feel confident your Salesforce tests can actually catch bugs. Github link in the comments. Tell us how your org's tests hold up. We'd love your feedback as we keep building.

55

6 Comments

Two buyers. Same flat. Two different prices. In India’s resale market, pricing often depends on who you speak to, not what the home is actually worth. At Truva, we decided early that if trust is going to exist at scale, pricing cannot be subjective. So we built TruIQ, our unit-level valuation engine. 🧠 TruIQ runs on 50,000+ verified resale transaction datapoints: past transactions, registry records, demand signals, occupancy status, floor plans, unit-level attributes, and hyperlocal trends. This enables something fundamentally different: • Pricing at the unit level, not just the building level • Accounting for real-world differences between “identical” homes • Grounding price recommendations in comparable transactions, not sentiment Last quarter, a seller challenged the price we recommended. Instead of negotiating, we opened the data. 🔓 We walked them through comparable unit transactions and showed exactly how we arrived at the number. There was no back-and-forth. They didn’t negotiate. They understood. And that difference matters. Today, our homes sell in 45 days on average. The broader resale market often takes 6 months or more. Fair pricing isn’t about convincing people. It’s about showing the math. 📊 Proud to work alongside Arun Koushik, Shreshth Gupta, Amulya Jain, Dishit Jain, Kamil Dehliwala, and the wider product, data, and engineering team behind TruIQ.

71

3 Comments

AI is slowly becoming part of daily life, not just office work. Google ’s AI platform Gemini has reportedly signed a three-year sponsorship deal with the IPL worth around 270 crore. It’s a big step, not only for cricket, but for how AI companies want to be present in India. Last year, ChatGPT sponsored the Women’s Premier League. Now Gemini entering the IPL shows something clear, AI brands are no longer staying in the background. They want to be where people already spend time and attention. Cricket in India isn’t just a game. It’s where millions of people gather every season. For AI platforms, this isn’t about banners or ads alone. It’s about becoming familiar and useful in everyday life. We’re moving from “AI is new” to simply “AI is here.” The real work now is making it practical, helpful, and easy for people to use. LinkedIn LinkedIn for Marketing LinkedIn Guide to Networking LinkedIn Learning

7

1 Comment

Google no longer provides free dinner to its employees on Friday in Bengaluru. This is nothing but cost-cutting. Google has been scaling back office perks at multiple locations since 2023, including reduced complimentary meals and micro kitchen services. With billions of dollars invested in LLMs, expect more cost-cutting measures. This is a reality check for job seekers: evaluate total compensation carefully, because what companies advertise as 'perks' today might disappear tomorrow when the market shifts. #Google #SDE

241

30 Comments

Yugabyte pays around 35+ LPA, and 50+ LPA CTC for more than 1+ years of experience to a software engineer 𝗥𝗼𝘂𝗻𝗱 1 – 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗥𝗼𝘂𝗻𝗱: 𝗤𝘂𝗲𝘀𝘁𝗶𝗼𝗻: 𝗗𝗲𝘀𝗶𝗴𝗻 𝗮 𝗛𝗶𝘁 𝗖𝗼𝘂𝗻𝘁𝗲𝗿. Focus: Ability to handle high-concurrency write requests and memory-efficient data storage. Discussion involved using circular buffers or buckets to manage time-based counts within a specific window. Follow-up: Scaling the counter to handle millions of requests per second and ensuring accuracy in a distributed environment. 𝗥𝗼𝘂𝗻𝗱 2 – 𝗖𝗼𝗻𝗰𝘂𝗿𝗿𝗲𝗻𝗰𝘆 & 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝗦𝗼𝗹𝘃𝗶𝗻𝗴: Technical Deep Dive: Basic questions on concurrency, thread lifecycle, and race conditions. Implementation: Design and implement a word auto-complete suggestion feature for keyboards. Approach: Utilized a Trie (Prefix Tree) data structure to store words and provided an efficient search algorithm for prefix matching. Follow-up: Optimizing the Trie for memory and adding weightage/frequency to suggestions. 𝗥𝗼𝘂𝗻𝗱 3 – 𝗔𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺𝘀 & 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝗟𝗼𝗴𝗶𝗰: DSA Question: Find the median of two sorted arrays. System Logic: Strategy for zero-downtime key rotation in a Master-Slave architecture. Problem Scenario: Master nodes SSH into slave nodes using a private key; slave nodes verify using a public key. 𝗥𝗼𝘂𝗻𝗱 4 – 𝗕𝗲𝗵𝗮𝘃𝗶𝗼𝗿𝗮𝗹 𝗥𝗼𝘂𝗻𝗱: Discussion: Deep dive into past projects and technical challenges faced in the last 4 years. Focus: Conflict resolution within a team, handling tight deadlines, and the rationale behind choosing Yugabyte.

396

6 Comments

🚨 Big update for India! Google just launched on-device AI scam protection in India. - ✅ Real-time scam alerts during calls from unknown numbers ✅ Privacy-first — analysis happens on your phone, nothing is sent to Google ✅ Warns you during screen-sharing if a payment app is opened ✅ One-tap option to end the call + stop screen sharing instantly ✅ Partners: Google Pay, Paytm & Navi ✅ New secure login tech: SIM-based ePNV (no OTP needed) India is now a testing ground for smarter, safer AI. Much needed with rising digital fraud. #AI #Google #DigitalSafety #IndiaTech #CyberSecurity #DigitalIndia #scam

6