Rahma Javed

Midjourney runs on models they own. Their inference costs drop every quarter. Their moat compounds every interaction. Your enterprise runs on Claude and GPT. Your costs go up with usage. Your moat is zero. This isn't a criticism — it's the default. Most AI deployments are subscriptions, not assets. You're renting intelligence, not building it. There's a different architecture. I call it the SLM Flywheel: → Deploy a small model trained on YOUR domain data → Collect production signals from every interaction → Detect when the model drifts → Retrain automatically → Redeploy smarter than before Three layers. One proprietary moat. Knowledge SLM: trained on your documents, transcripts, policies — not retrieving from them at runtime. It understands your domain. Operational SLM: captures how your best agents make decisions and execute workflows. Your institutional expertise becomes a model. Autonomous Retraining: monitors drift and retrains without you. Your January model doesn't go stale by March. The result: 5–50× cheaper than frontier APIs. Sub-200ms latency. A model no competitor can replicate — because they don't have your data. The enterprises that start the flywheel today will be impossible to compete with in 3 years. Where is your enterprise on this curve? #EnterpriseAI #SLM #AIStrategy #MachineLearning #DeploymentScience #Uniphore

38

11 Comments

When you equip a top-tier engineer with AI coding agents, their output doesn't bump 20%. It scales by an order of magnitude. If that person's capacity is suddenly worth $2-3M a year — pulling them into a 12-person cross-functional alignment sync is not just annoying. It's a catastrophic destruction of enterprise value. And yet. Every company does it. Every week. To their best people. Here's why: The math of coordination has broken. In the pre-AI world, more people = more output. That was the deal. So you accepted the coordination tax: 5 people meant 10 communication pathways. 10 people meant 45. 20 people meant 190. Meetings were the price of human volume. AI made the volume free. But nobody recalculated the tax. Your best engineers are still sitting in the same meetings, attending the same syncs, writing the same status updates. Except now each hour in that meeting room costs 10x what it did two years ago — because that's what their AI-amplified output is worth when they're NOT in the meeting. It gets worse. When you put 20 people on a project and give them all AI tools, they generate massive amounts of code, tickets, and ideas at machine speed. But the team is too large to share a coherent mental model. The AI outputs start contradicting each other. So leadership schedules more meetings to verify. Those meetings generate more decisions. Those decisions generate more AI tasks. Welcome to the agentic tarpit. The fix isn't better meetings. It's fewer people. A five-person strike team doesn't need status meetings because there's nowhere to hide. Everyone holds the system architecture in their head. You point five people at an ambitious goal, give them autonomy, and get out of their way. That's the team I want to be embedded in. Not managing a department. Not running alignment syncs. Building — with four other people who have taste, judgement, and the tools to move at machine speed. The question every leader should be asking right now: are your best people building, or are they sitting in meetings about building? #FieldAI #AIinProduction #BoardToBot #AgenticCommerce #BuildInPublic #OpenToWork

8

1 Comment

Thrilled to share my latest article: "The Mixed-Initiative Interface: Designing Control Handoffs Between Humans and AI." In a world where AI handles more decisions, the real UX frontier is the seamless transfer of control—ensuring humans can step in without losing context, state, or accountability. We're moving beyond rigid autonomy to dynamic systems where handoffs are graceful and empowering. Key insights include: - The Handoff Problem: Mixed-initiative systems allow bidirectional control shifts; the challenge is preserving state during transitions to avoid confusion. - Decision Ledger: An append-only log of actions, confidence scores, and reasoning to brief humans on AI's prior steps. - Non-Destructive Overrides: Use proposal queues for AI suggestions that can be reviewed or modified without halting workflows. - Calibrating Confidence: Dynamic thresholds based on consequence, reversibility, and confidence—err toward escalation to build trust. - Reversibility and Intervention Windows: Leverage event sourcing for undos and strategic delays to enable human oversight in high-stakes scenarios. This is essential for anyone building AI-powered products: think pricing optimization, fraud detection, or any domain blending human judgment with machine efficiency. Does this align with challenges you're facing in your work? Read the full piece here: https://lnkd.in/dX_y7YMa

5

Had a great conversation with Amir_ Raminfar, Software Engineering Manager at Netflix, about where engineering teams are heading next. Teams are already moving faster than their processes can handle. That gap is creating a new kind of pressure most orgs aren’t ready for yet. Full interview drops on Tuesday by 8 AM EDT. Subscribe to Exec Engineering so you don’t miss it (link in comments ����) Thank you Amir for sharing your insights!

6

1 Comment

63% of companies are now using AI for most coding. But many engineering leaders are still stuck stitching together dashboards just to understand what’s happening. That gap is what we’re tackling at Jellyfish. Today, we're introducing a major update to Jellyfish Assistant – an AI-powered conversational tool that turns engineering data into instant, actionable insights. Instead of digging through reports, you can ask: • Where are we slipping on delivery? • How is effort aligning to our priorities? • Which teams are at risk? …and get answers with context and recommendations. You can learn more about what a smarter Jellyfish Assistant means for you here: https://lnkd.in/eDd_AGzs

18

Mastercard + PayPal: Bringing Agentic AI Payments Mainstream The shift from concept to reality is happening faster than anyone expected. Today, Mastercard announced that its new Agent Pay technology will be integrated directly into the PayPal wallet - enabling consumers to make purchases guided by Agents, while Mastercard provides the trusted rails beneath the surface. This collaboration connects PayPal’s millions of merchants and co-branded cards to a shared agentic infrastructure; powered by trust, tokenization, and security at scale. 🎙️Mastercard’s Chief Digital Officer, Pablo Fourez, called it “a paradigm shift bigger than the move to mobile.” Together, the two companies are solving the hardest problems in agentic payments: ✅Agent verification and consent capture ✅Dispute management and interoperability ✅Real-time, verifiable credentialing The first wave of agentic shopping experiences is expected to roll out during the 2025 holiday season, marking the start of a new era where your AI doesn’t just recommend what to buy… it completes the transaction for you. This is what the future of commerce looks like - intelligent, trusted, and seamless. 🔗 More here: https://lnkd.in/gHiXq6jn

57

1 Comment

AWS just dropped $100 million on AI agent development. But the real opportunity isn't enterprise workflows. While big tech builds AI for Fortune 500 companies, B2B founders are still drowning in busywork that pulls them away from building. Founders need AI agents for everyday work - not just enterprise workflows. Things like finding real conversations your audience cares about, drafting personalized outreach, or turning customer feedback into content. Enterprise gets AI for complex sales automation. But who's building agents that handle the small, repetitive tasks that steal founders' time from product development? This week, Sorad surfaced discussions about the latest OpenAI announcement and turned it into a LinkedIn post series that led to over 2k+ growing impressions. Comment PROMPT with your website link and I will DM you the free strategy Sorad created for you. #AIAgents #ContentStrategy #B2BMarketing

6

2 Comments

Payments have always worked on a simple idea: A person decides to buy something. That is what the whole system is built around. Now that is starting to change. With things like Visa’s “Agentic Ready” work, we are moving toward a world where software does not just help you decide… It can actually make the purchase. That is a pretty big shift. Because payments were never just about moving money. They are about knowing: 1. Who made the decision 2. Why it was made 3. Who is responsible When a person clicks “buy,” that is clear. When an AI agent does it, it is not so obvious. Who owns that decision? How do you know it was authorized correctly? What happens if something goes wrong? This is not just a new feature. It changes how the system needs to think. Instead of: Person → Payment It becomes: Person → Rules → AI → Payment And suddenly, trust looks different. We are no longer just verifying people. We are starting to verify software acting on their behalf. That is going to be interesting to watch. Ref: https://lnkd.in/gajP2gab

24

Here's my take on hiring and getting hired: Evaluating problem solving skills, emotional intelligence, empathy, curiosity, trustworthiness, and creativity is not an easy task which requires actually talking to someone, while possessing those same qualities. It is a vibe, not a grind. If you're lacking elements of the vibe you will not perceive their presence or absence in the other.

30

Stripe DBaaS serves >5,000,000 QPS from product applications 10,000+ distinct query, PetaBytes data in 5,000+ collections, 2,000+ db shards Problem: Off-the-shelf Database as a Service (DBaaS) isn't available to meet Highest standards of availability, durability and performance. Support for multi tenancy with enforced quotas, horizontal scalability with sharding and Strong security through enforcement of authorization policies Solution: Stripe build DocDB with MongoDB as the underlying storage engine with online data migrations at its core Architecture 1. Applications access data through a fleet of database proxy servers, developed in-house in Go to enforce reliability, scalability, admission control, and access control 2. To horizontally scale, 1000s of database shards house a small chunk of the cumulative data 3. Application sends a query to a database proxy server (PS), it parses the query, routes it to one or more shards, combine results from shards and returns them back to the application 4. PS rely on a chunk metadata service that maps chunks to database shards. To look up the shards for a given query 5. Change events from writes to database are transported to streaming software systems, and eventually archived in an object store via CDC pipeline 6. Product teams use database control plane to provision a logical container (logical database) for their data housing one or more DocDB collections 7. Data in collections is distributed across several databases (physical databases) hosting a small chunk of the collection 8. Physical databases on DocDB live on shards deployed as replica sets that comprise a primary node and several secondary nodes with replication and automated failover 🤔 over to you, how to migrate data across database shards in a client-transparent manner with zero downtime? Let me know in the comments! 👇 --- how to showcase full range of system design skills in 60 minutes interview?     Join 2200+ subscribers for a FREE 7-day email course 👉 systemdesign(dot)engineer (link in comments)

51

11 Comments

Your CTO shipping code might feel reassuring. At scale, it’s often a sign the system isn’t working. We often celebrate a hands-on CTO. It feels safe. Execution is visible. Code is shipping. But there’s a quiet trade-off most scale-stage companies don’t notice immediately. Not because coding is bad. But because roadmap, strategy, and vision conversations slowly disappear. One early warning sign I’ve seen in scaling companies: When a CTO spends more time coding or reviewing PRs than leading discussions on where the system is going. The symptoms are subtle: • Senior leaders evaluated like individual contributors • Architecture evolving reactively instead of intentionally • Teams busy, but not aligned • Strategy quietly becoming “someone else’s problem” I’ve seen how this shapes behavior. When leaders are evaluated like ICs, they stop acting like leaders. They optimize for visibility, not outcomes. Marty Cagan captured this shift well: “Leaders are responsible for results, not for doing the work themselves.” Ben Horowitz framed it from a scaling lens: “The most important skill for an executive is the ability to get the right things done through other people.” The strongest CTOs I’ve worked with at scale weren’t the most visible coders. They were the ones creating clarity—around direction, trade-offs, and leadership beyond themselves. Coding can be a powerful tool—especially early on. But when IC-level contribution becomes the primary yardstick for leadership, the hidden costs show up later: plateaus, rewrites, attrition, missed windows. A question for CTOs, founders, and investors: At what point in your company’s growth did staying hands-on stop helping—and start limiting scale? #CTO #TechLeadership #ScalingStartups #EngineeringLeadership #StartupLeadership

50

9 Comments

🚀 Agentic Commerce - the next leap in how AI and payments converge. Google just unveiled its Agent Payments Protocol (#AP2), while Stripe introduced Agentic Commerce Solutions and an open standard (#ACP) built with OpenAI. In simple terms: 🤖 AI agents can now buy, subscribe, and transact on your behalf — securely and transparently. 🔐 Using scoped, revocable payment tokens (SPTs) instead of raw credentials. 🌍 All powered by a unified open spec: agenticcommerce.dev Why it matters: • The checkout moves inside AI experiences — users shop where they discover. • Merchants get one protocol for multiple AI platforms. • New frontiers for trust, identity, and fraud protection emerge. Agentic commerce could redefine how digital ecosystems handle AI-driven purchasing — from “AI discovery” to “AI buying.” Stripe ACP - https://lnkd.in/d5hbbhWQ ACP - https://lnkd.in/daZbpVjy OpenAI - https://lnkd.in/dhMnFf-C Stripe Docs - https://lnkd.in/dFK6MQ7w #AgenticCommerce #Fintech #AI #Payments #Stripe #Google #OpenAI #FutureOfCommerce

25

2 Comments

My next guest, Eno (co-founder of Factory), built this insane product management AI skill that blew my mind. It includes: → Product principles and positioning → What an 11-star experience looks like → PRD templates and review rubrics → Prioritization frameworks ...and more It's 700+ words and I'll admit it made me a little worried about my profession 😅 Watch the full interview now: https://lnkd.in/gapCAhf4

279

10 Comments

POS UI Extensions can now run without network access in Shopify POS 11.0. Why this matters: Retail doesn't stop when the internet does. Pop-up shops, outdoor markets, large retail floors with spotty Wi-Fi. These are real environments where merchants need their tools to keep working. Now, third-party extensions built on our platform can do exactly that. It's an opt-in feature, one line in your extension config, and your extension keeps functioning using locally available data and APIs when connectivity drops. No API update required! This is the kind of infrastructure work that doesn't always get the spotlight but makes the platform meaningfully more reliable for the developers building on it and the merchants depending on it. Changelog: https://lnkd.in/gQA3UrHv Docs: https://lnkd.in/g3bSw7fu #Shopify #POS #RetailTech #Extensibility

55

8 Comments

No existing medium or large company is going to get a large multiplicative factor (2x, 5x, 10x) of execution speed by giving their team access to AI tools. Example - you give your devs full access to Claude. Your Product Managers have ChatPRD. Your execs have agents to summarize their emails and decks. Yet you’re still moving the same speed more or less. Why? Your communication is still MxN. That PRD? Bob’s out until next week and wants to be there for the review. Then you need to get it on the VP’s calendar for an approval stamp. That feature? It’s across 3 teams and needs at least one Pinciple Engineer review. Did you hear that one of the teams is heads down on a high priority CEO ask? Good luck getting that into their roadmap. Who wins? The small companies. The startups. They can maximize the leverage of all these tools and they don’t have the communication over head.

27

1 Comment

Every wave of innovation meets resistance — until it reshapes the world. Years ago, a senior financial executive told me: “We can’t stop crypto with innovation, so we’ll have to stop it with legislation.” That mindset shows up every time technology challenges entrenched systems — from the power loom to the internet, from crypto to AI. Resistance rarely comes from a lack of vision; it comes from fear of losing control. After 10+ years at Amazon leading teams behind Alexa’s mobile, web, and device platforms, I’ve seen this tension up close. Our work has always lived at the intersection of hardware, AI, and customer experience — where milliseconds, sensors, and speech models have to work in concert across billions of devices. It’s one thing to build AI in the cloud; it’s another to make it feel instant and human in your pocket, car, or living room. That’s where the real future is being built — at the edge. Where AI isn’t an abstract model in a datacenter, but an adaptive layer of intelligence running directly on devices, learning in context, and acting locally even when the cloud is out of reach. This shift demands a different kind of engineering — one that blends distributed systems, privacy, and machine learning into something customers can trust. It’s not just about scale; it’s about responsibility. Generative AI gets a lot of attention for “hallucinations,” but in many ways, that creativity is what fuels innovation. The challenge isn’t to eliminate it, but to channel it — designing systems that balance imagination with truth, autonomy with accountability, and human intuition with machine precision. I’ve spent my career building platforms that bridge those worlds: from large-scale distributed file systems to on-device voice intelligence to now helping shape Alexa’s next-generation experiences across phones, wearables, and connected ecosystems. Every time technology forces an evolution, the same question emerges: Are we building guardrails that enable transformation, or barriers that delay it? We’re standing on the edge of the next great inflection point. And I, for one, plan to keep building.

12

Today’s new research from Harvard Business Review highlights something many of us feel intuitively: AI doesn’t actually reduce work - it “intensifies” it (e.g., enables increased output). In an eight-month study, employees using generative AI sped through tasks, expanded their output, and ended up juggling more, not less. Think about the printing press. Before Johannes Gutenberg, books were laboriously hand-copied - slow, expensive, and niche. The printing press didn’t reduce the need for writing or reading; it exploded literature production, fueled new industries, and changed expectations about who could share and access ideas. People didn’t work less they produced and consumed far more. There’s an important lesson for today in that. https://lnkd.in/enuxt6Cs

31

1 Comment

In one of our enterprise deployments in an Abu Dhabi bank, a single user account is now driving the prompt volume of what used to be 20 analysts. That user is orchestrating agents that run overnight. The seat count looks the same on the invoice. The actual usage looks nothing like it. Engineering teams building SaaS need to think hard about this. If your billing, rate limiting, and capacity planning are all anchored to humans, you're going to get surprised on both sides: customers paying for seats nobody logs into, and infrastructure melting under agent traffic from the seats they do. We rebuilt our metering around prompts, tokens, and policy decisions instead of users. Identity still matters for audit and access control. It just stopped being the unit of value. The SaaS pricing model that wins the next five years is the one that prices the work, not the worker.

8