@vasuman 3/3: 18: unapplied credit memos. Vendor credits sit unused while invoices get paid in full. Surface open credits and net the approved ones against the next payment run, with a check so nothing is applied twice.
@vasuman 2/3: 17: early-pay discount windows. Approval time quietly burns the discount date. Track the window on each invoice and escalate when the approval path is about to cost the discount. That is cash lost to process, not negotiated terms.
What if device owners like Google and Apple ship a default interface like Instinct/Muse? They already own distribution, can build strong UI, Google has the models/compute, and Apple can always rent them. Instinct’s and even muse messaging-layer play starts looking pretty fragile.
instinct will be a textbook example of an AI boom failure.
for a consumer AI agent to win, it needs three things:
distribution
great UI
compute (latency + memory)
instinct found a clever distribution hack: live inside messaging platforms so users don't have to download another app.
but that comes at a cost. you don't own the platform, and therefore you don't control the UI.
one of muse's biggest advantages is its ability to offer a dynamic, interactive interface beyond the traditional text-based chatbot. instinct simply can't replicate that unless the platforms it relies on allow it to.
distribution and UI go hand in hand. instinct optimized for distribution by sacrificing control over the user experience. eventually, it'll pay the price.
then there's compute.
muse has the infrastructure and resources to deliver better latency and memory. instinct had to raise billions just to compete on that front.
the fact that investors continued funding instinct after muse launched is genuinely baffling to me.
people in sf might be willing to try ten different AI agents. the average consumer won't.
if someone's first exposure to AI agents is muse through instagram, threads, or whatsapp, what reason do they have to seek out instinct?
of course, there's a bigger question of whether consumer agents will succeed at all.
an agent-first world would fundamentally reshape consumer commerce. brands like amazon will be reduced to fulfillment providers, becoming one of many interchangeable options rather than the destination itself.
but if that shift does happen, i don't see instinct being the winner.
it chose to build on someone else's distribution, without owning the interface or the infrastructure.
and it's competing against a company that owns all three.
@rodhowarth Core infra & systems engineers will enjoy AI most—they build the real foundations. Customer-facing app builders will have it rough. Eventually non-engineers and semi-engineers will use the toolsets built by actual engineers to ship those apps.
@HungLee Both can be true. The real junior employment drop in tech + big tech growing without adding headcount. a16z is looking at job postings and how the AI-job-loss panic cooled. Postings ≠ easy junior hires. Market favors experienced people; pure entry-level is still tough.
@levie The job was never to write code. It was always to make the computer do a task. Languages evolved from machine code to C/C++. Tools evolved from IDEs like Visual Studio to agents. You remain responsible for the task. The toolset keeps getting better.
@Grady_Booch@unclebobmartin That’s like saying a driver’s job isn’t transporting boxes just because they now drive a truck with 10 instead of carrying one by hand. You’re still responsible for getting the boxes from A to B — you just have better tools now.
@gbaheti Win US SMEs first, then sell to US enterprises. Then come to India and sell to Indian enterprises. It's relatively easier to sell and scale in the US, and that proof helps you win in India.
Engineering depth and domain expertise are different skills. Finding both in one person is hard, even with AI. I'd back engineers building better tools than Lovable, and domain experts using them to build better solutions.
I expect this trend to really grow.
Multidisciplinary engineers, especially with high customer empathy and product sense will be in high demand
https://t.co/xQaDP4rhNW
An interface is never the Figma/Sketch file. It is how a human and a machine communicate. Screen, click and touch were one form factor; reasoning models add words and voice. Designers don't disappear, they move from drawing screens to designing the interaction across all of them.
Interfaces are not going anywhere. Agents will co-exist and interfaces will evolve to accommodate them for new HCI paradigms.
Talking is sometimes more useful than tapping or typing. Reading a chat with a chatbot is sometimes more interesting than selecting from an endless grid. Not always
Interfaces are not going anywhere. Agents will co-exist and interfaces will evolve to accommodate them for new HCI paradigms.
Talking is sometimes more useful than tapping or typing. Reading a chat with a chatbot is sometimes more interesting than selecting from an endless grid. Not always
@levie Procurement is a good example. Extracting information from documents, reading contracts and redlining were always manual because the technology did not support them. Now it does. The constraint shifts from writing code to knowing the process well enough to specify it.
@jasonlk Guardrails belong around the tool call. In what we build, outbound email needs manual approval past 10 messages to one person, agents have no delete tool at all, and write tools stop working when behaviour looks abnormal, like a spike in edits.
@harshilmathur Now imagine you have a robot at home. You tell it to bake a cake you saw on MasterChef, and it makes one using that oven. You just want a cake.
@officialepfo@epforokandivali@LabourMinistry Two grievances for my PF transfer-in to Kandivali East, both answered with a two-week wait. It has been pending for 3 months and is still not credited. Please resolve it and confirm the credit. Grievance 31321.
@officialepfo@epforokandivali My PF transfer-in to Kandivali East pending for 3 months. I raised one grievance and was told to wait two weeks. Nothing happened. I raised another, 31321. Same answer. This is urgent. Please get the transfer credited, not just a status update.
@GovindK37543714@officialepfo@socialepfo@LabourMinistry I have a similar problem. My PF transfer-in to Kandivali East pending for 3 months. Two grievances raised, latest 31321. Only told to wait two weeks each time. Please get the transfer credited, not just a status update. https://t.co/F3INKTuee7
@officialepfo@epforokandivali My PF transfer-in to Kandivali East pending for 3 months. I raised one grievance and was told to wait two weeks. Nothing happened. I raised another, 31321. Same answer. This is urgent. Please get the transfer credited, not just a status update.
@nikitabier Technical moats were never the real differentiator anyway.
It’s always been domain + distribution.
You need to understand the problem better than anyone else — and know exactly where your customers are looking to solve it.