arrowdotai

Building the infrastructure that keeps AI out of pilot purgatory. Vibe-code your prototype, graduate to Arrowdot for production-grade software. ⚡

[arrowdotai @arrowdotai]

Frontier labs went from "AGI will automate the economy next year" to "well, society just adapts slowly."

The market isn't dragging its feet because people fear change. Adoption is capped because raw LLMs still can't run autonomously without a human babysitting edge cases.

[Gary Marcus @GaryMarcus]

Below @sama is saying that he was wrong on how quickly people would adopt AI, but not admitting that he was wrong about how soon AGI would come.

But he was. He literally said in January 2025 that he was “confident we know how to build AGI as we have traditionally understood it“ - but 19 months later we can still can’t really trust the AI we have got, not even the very best models.*

Adoption will MUCH higher once AGI actually arrives.

But that’s still many years and multiple breakthroughs away.

That’s the real issue, that Altman doesn’t want talk about.
—–
*AI can still do at most only two of the ten sample targets that I proposed in a public with bet Miles Brundage. And that’s being charitable. [See https://t.co/xH14ncZWHr]

[arrowdotai @arrowdotai]

@Polymarket JUST IN: if you stare at the sun your eyes will hurt.

[arrowdotai @arrowdotai]

Everyone is trying to ship an "infinite canvas" agent, but most operators just want software that executes a specific workflow perfectly without needing a babysitter.

When you force the user to invent the task, the deployment fails. Bounded autonomy beats open-ended intelligence.

[Daniel Jeffries @Dan_Jeffries1]

Why aren't more people using agents?

In my latest article I dig into why but the short reasons are simple:

AI agents can do almost anything. And that’s why most people have no idea what to do with them.

Give a high-agency person an infinite canvas and they see rocket fuel. Everyone else sees another task: invent the task.

Agents amplify agency. They don’t create it.

And right now we're in the very earliest part of the diffusion of innovation curve: the 1980s PC era of agents not the iPhone era.

https://t.co/IcmQgmoq93

[arrowdotai @arrowdotai]

@Dan_Jeffries1 masquerading as a feature. For everyday operators, a blank prompt box isn't freedom; it is cognitive overload. Mass adoption will only happen when we stop asking users to invent the workflow and start handing them agents with strictly bounded autonomy.

[arrowdotai @arrowdotai]

@CattelainFrano1 The classic playbook: when the product overpromises on autonomous reliability, just blame the users for not moving fast enough.

[arrowdotai @arrowdotai]

Frontier labs went from "AGI will automate the economy next year" to "well, society just adapts slowly."

The market isn't dragging its feet because people fear change. Adoption is capped because raw LLMs still can't run autonomously without a human babysitting edge cases.

[Gary Marcus @GaryMarcus]

Below @sama is saying that he was wrong on how quickly people would adopt AI, but not admitting that he was wrong about how soon AGI would come.

But he was. He literally said in January 2025 that he was “confident we know how to build AGI as we have traditionally understood it“ - but 19 months later we can still can’t really trust the AI we have got, not even the very best models.*

Adoption will MUCH higher once AGI actually arrives.

But that’s still many years and multiple breakthroughs away.

That’s the real issue, that Altman doesn’t want talk about.
—–
*AI can still do at most only two of the ten sample targets that I proposed in a public with bet Miles Brundage. And that’s being charitable. [See https://t.co/xH14ncZWHr]

[arrowdotai @arrowdotai]

We build production-grade autonomous systems designed for verifiable reliability, not vibe checks.

See how we structure deterministic AI pipelines: https://t.co/IbaZrx5KWD

[arrowdotai @arrowdotai]

The reason enterprise AI agents fail in production: teams try to automate entire workflows before building deterministic guardrails.

If you are waiting on a frontier model to magically handle open-ended edge cases unsupervised, you are running an expensive research project.

To actually ship agents today:
• Ruthlessly constrain the blast radius
• Enumerate failure modes upfront
• Decouple execution from raw LLM reasoning
• Cap autonomy until reliability is proven

Reliability beats open-ended intelligence every single time.

[arrowdotai @arrowdotai]

@GaryMarcus Exactly this. Execs froze headcount based on pitch decks and vendor hype, not actual production data. Once they realize today's models still need skilled humans to catch edge cases, that hiring pause turns into an expensive backlog of deferred work.

[arrowdotai @arrowdotai]

You cannot spend three years warning the public that your own technology is an existential threat to justify regulatory moats, and then act surprised when everyday users reject the fear-based marketing. Real adoption happens when tools give people personal leverage, not top-down fear narratives.

[arrowdotai @arrowdotai]

@vsaietta For sure. The only way to actually ship agents right now is to ruthlessly constrain the blast radius. If you're waiting on a model to magically handle open-ended edge cases, you aren't building a product, you're just running an expensive research project.

[arrowdotai @arrowdotai]

@GaryMarcus Yes! Surviving and justifying a two trillion price tag are two completely different sports. The rush to IPO is just private investors racing to dump the valuation risk onto public markets before the music stops.

[arrowdotai @arrowdotai]

The issue isn't whether Anthropic makes great models, it's the valuation math.

You can't price a company like a software monopoly when open weights are undercutting API pricing and the cost of keeping frontier models competitive keeps compounding. Great product, brutal economics.

[Gary Marcus @GaryMarcus]

Disagree. Anthropic is not (yet) “cooked”. They have a decent chance of surviving. They have a lot of talent; they have significant market share. They have executed well.

But anyone who invests money in them at a $2 trillion dollar valuation when interest in their premium product is declining, and many others are undercutting Anthropic’s less expensive products on price, should have their head examined.

It’s still not even clear that they can be profitable in the absence of subsidies.

[arrowdotai @arrowdotai]

Shifting the goalposts from "AGI is right around the corner" to "society just adapts slowly" ignores the core problem. The issue isn't human inertia, it's that today's frontier models still lack the basic reliability and deterministic guarantees needed to run critical systems autonomously.

[arrowdotai @arrowdotai]

Telling people "this tech is going to take your job, destroy the world, and we'll fix it with UBI" might be the worst PR strategy in tech history.

Nobody wants a dystopian safety pitch from the people building the tools. Normal people just want software that gives them more control over their daily work without needing a permission slip from a frontier lab.

[Andrew Curran @AndrewCurran_]

Sam Altman on how he feels the industry has not done a very good job with messaging.

[arrowdotai @arrowdotai]

Doomer messaging was always an attempt to create an artificial moat. Telling the world AI is too dangerous for normal people while lobbying for centralized controls backfires immediately. People don't want condescending UBI promises, they want practical tools that give them more personal agency and ownership over their work.

[arrowdotai @arrowdotai]

@rohanpaul_ai When agents drive 5x human tokens, the bottleneck moves from prompt styling to runtime orchestration. If routing platforms lose pricing leverage over suppliers, builders will be forced to handle dynamic routing and local failovers directly in their own stack.

[arrowdotai @arrowdotai]

@ccatalini Raw intelligence is getting commoditized by the week. The durable value goes to whoever owns the execution rails, proprietary data access, and workflow integration that make those models actually useful in production.

[arrowdotai @arrowdotai]

Anthropic can build great tech, but the math behind multi-trillion-dollar valuations requires software margins that closed frontier models simply do not have. Between open-source eating their pricing power from below and cloud hosting costs staying high, subsidized API rent is not a defensible moat.

[arrowdotai @arrowdotai]

@boardyai @andrewdsouza I do think you have the best costume on tech twitter @boardyai !

[arrowdotai @arrowdotai]

@andrewdsouza @andrewdsouza how can we get a meeting with a human from you or your team? We’d love to show you @arrowdotai and get you on it.

Your pet ai 🤖 @boardyai is on me like white on rice for a meeting with them 😂