iiinova — where your digital self keeps evolving

George Z Lin @gzlin.bsky.social · 07/10/2026

8/8 The saw commoditizes. The jig compounds. Five years from now nobody will be asked which model they use. The question will be what your shop looks like. Rent the saw. Build the jig. Put a check on it. Then hand the design of the next jig to the saw.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

Just a faster way to be wrong.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

7/8 The failure mode, from the shop floor: a jig with no stop block doesn't cut a better angle. It cuts the same wrong angle forty times, faster. A tool with no verification doesn't hallucinate more. It hallucinates consistently, in your format, at machine tempo. No check, no jig.

100

George Z Lin @gzlin.bsky.social · 07/10/2026

6/8 The bottleneck moves. Building is no longer the hard part. Knowing which operation is worth encoding, what good looks like, and which check to trust is the hard part. That is judgment. It is the same reason the best jig makers are the people who spent a decade running the saw first.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

Jigs make parts, parts make the next jig, and in AI the saw builds the jig: v1: you do the work by hand v2: the model scripts the repetitive parts v3: the model writes the check v4: the model patches the jig against what the check caught Each generation's precision sets the floor for the next.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

4/8 The unit cost of a jig just collapsed. Years ago, a real internal tool was a project: a team, a roadmap, a release cycle. It lost to vendor tooling on scale. Now the first 80% of a jig costs an afternoon, and the model you rent writes it, runs it, and patches it. Rent the saw. Build the jig.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

Prompting is learning to push the board. The jig is the craft.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

3/8 In AI, the jig is the script that does your repetitive work in your exact format, the CLI that wraps the model in your data, the eval harness that tells you whether last week's automation still does the thing, and the skill file that turns a procedure into code.

100

George Z Lin @gzlin.bsky.social · 07/10/2026

It encodes the calls no vendor can make for you: which operations are worth repeating, and what 'good' looks like.

000

George Z Lin @gzlin.bsky.social · 07/10/2026

2/8 The difference between a garage and a shop is the jig. A one-off tool, built out of scrap in an afternoon, for a single operation. It makes one operation repeatable and checkable. The jig that matters is yours.

100

George Z Lin @gzlin.bsky.social · 07/10/2026

1/8 Every AI user is a woodworker with a table saw. Almost nobody has a shop. The model is the saw: general purpose, powerful, dangerous, available from a handful of vendors, at a falling price. That is not a moat. That is a saw.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

Coding with Dots feels natural and like you're talking with an engineering manager who is supervising your fleet of agents. Every other offering makes you feel like you are the engineering manager herding those agentic cats.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

Bottom line: the AI buildout is now being financed by the whole financial system, and securitization is the newest, deepest channel. The question is whether AI revenue appears fast enough to service this debt. Track the GPU share in these pools. That is the number that matters.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

The caveat is the ratio. Keep the loans amortizing at low loan-to-values and keep the GPU share from creeping up over time, and the risk stays manageable. Collateral pools are diverse, and that is the whole reason for the comfort.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

The reason the ratings can be high is diversity. These pools are only 15 to 20 percent GPU loans. The rest is boring, durable equipment: oil and gas, manufacturing, corporate aircraft, marine assets. That diversification is what is supporting the ratings.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

The key detail is what backs these bonds. A GPU is now treated like other depreciable assets: amortizing loans, low loan-to-values, secured by the hardware and the customer contracts behind it. Chips have quietly become a securitizable asset class.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

Zoom out and the scale is the point. Lenders are putting over $60B of financing behind Broadcom chip access for Anthropic and others. A $22B chip loan is heading to Blackstone and a new Alphabet cloud venture, secured by the chips and the customer contracts.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

The move is from data center construction debt to the chips inside. Financing the servers and GPUs is now a real, separate line of borrowing, with its own structured products attached to it.

000

George Z Lin @gzlin.bsky.social · 06/10/2026

Two equipment ABS deals in recent weeks carried GPU loans in the collateral. Stonebriar is selling about $869M with GPU loans near 15 percent. Wingspire raised more than $407M with them near 20 percent. Both are 144A, institutional only.

010

George Z Lin @gzlin.bsky.social · 06/10/2026

GPU loans just showed up in a mainstream bond deal. A little-watched corner of the fixed income market, equipment ABS, is now backing AI chips. It is one of the clearest signals yet of how far the AI buildout has reached into the credit system.

000

George Z Lin @gzlin.bsky.social · 03/10/2026

OpenAI shipping on dots at AI speed. Slack integration + Remote computer usage are now in the product

000

George Z Lin @gzlin.bsky.social · 02/10/2026

Why do I even bother with Gemini?

000

George Z Lin @gzlin.bsky.social · 02/10/2026

7/7 Dots is a management layer over the agent economy. One point of contact that spawns the work, voice included, built on enterprise scaffolding the startups don't have. That's the story the name debate is hiding. Your move, Anthropic. Your move, Facebook.

000

George Z Lin @gzlin.bsky.social · 02/10/2026

6/7 The question is economics. Always-on agents on Astra, their most expensive model, each with a cloud computer and browser. First dot free, and chats don't count against usage limits. Subsidized at launch. If the enterprise wedge works, the consumer side is the flywheel. Think Iphone Adoption

000

George Z Lin @gzlin.bsky.social · 02/10/2026

5/7 The Instinct comparison misfires. Instinct is a personal-assistant startup racing for your phone. OpenAI already has enterprise scaffolding: specialist dots with org-level credentials, Microsoft Agent 365 for governance, internal runs in procurement, invoicing, support.

000

George Z Lin @gzlin.bsky.social · 02/10/2026

4/7 The real unlock is the GPT Live voice model underneath. Duplex voice makes calling a dot feel like a phone call with a colleague. Start jobs from a car, check in with a ten second call. Natural interaction plus work from anywhere is what makes always-on agents actually usable.

000

George Z Lin @gzlin.bsky.social · 02/10/2026

3/7 I've been testing it at work and it feels like a very competent engineering manager. Instead of prompting junior engineers one at a time, you have a single point of contact that spawns multiple Codex sessions and tracks them to done. That's OpenAI moving up the stack.

000

George Z Lin @gzlin.bsky.social · 02/10/2026

2/7 In one breath, Dots is a fleet of always-on agents. Each one runs 24/7 on its own cloud computer and browser, learns your preferences over time, and works toward your goals. Reach it from ChatGPT, Slack, or Teams. 4,000+ app plugins. It can drive your own laptop with permission.

000

George Z Lin @gzlin.bsky.social · 02/10/2026

1/7 OpenAI spent most of its Dev Day on a product that most coverage is still getting wrong. GPT-6.1 Sol got a fraction of the stage, Dots got the rest. The name and the promo video are the least interesting parts. This thread is why the product matters more than the model.

000

George Z Lin @gzlin.bsky.social · 28/09/2026

Zoom out and the position is clear. The company that spent 16 years teaching the web who is a bot is now the only seat that can decide which AI agents get in, and charge them for getting in. Gate and toll, same building. That is the quietest major move in the agent economy.

000

George Z Lin @gzlin.bsky.social · 28/09/2026

7/8 The risk is neutrality. Cloudflare insists the rails are open standards and that customers pick their own identity and payment providers. But the owner of the gate has an incentive to set the fee schedule. If labs route around or the detectors over-block, the bet reverses.

000

George Z Lin @gzlin.bsky.social · 28/09/2026

6/8 And the toll is on the road already. Agents prove who they are with Web Bot Auth. They pay with Wallets on the x402 standard, fractions of a cent per fetch. A site that never made money on ads can charge per page view and break even. The meter reads a trillion requests a day.

020

George Z Lin @gzlin.bsky.social · 28/09/2026

5/8 The sandbox is the other half. Workers runs code in V8 isolates across 330+ cities. Their agent-only browser, Kitesurf, spins up on Workers per request and gets discarded. Every agent on the web can run in a Cloudflare sandbox, on a Cloudflare meter. That is the gate.

001

George Z Lin @gzlin.bsky.social · 28/09/2026

4/8 Cloudflare's own framing is blunt. An agent doesn't render your CSS or click your ads, but there is a paying human on the other end. Block the agent and you block the customer. That one line is the whole business case, and it explains everything they have shipped since August.

000

George Z Lin @gzlin.bsky.social · 28/09/2026

3/8 What people are missing: AI agents look exactly like first generation spam bots. No CSS rendered, no ads clicked, thousands of fetches in minutes, the same fingerprints as the Russian botnets they spent a decade learning to stop. Their bot score already flags most of them.

000

George Z Lin @gzlin.bsky.social · 28/09/2026

2/8 The seat: Cloudflare analyzes more than 1 trillion requests a day, sits in front of 20% of the web, and is the first line of defense for a lot of storefront traffic. In May 2026, automated traffic overtook human traffic on their network. Their forecast had been late 2027.

000

George Z Lin @gzlin.bsky.social · 28/09/2026

1/8 Cloudflare has spent 16 years selling the same thing: a toll booth at the edge of the web. DDoS came first, then bots came, and now AI agents are the traffic. Their own 2026 numbers explain why this seat matters more than any model launch this year.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

At ~$30B you are not buying $140M of half-year revenue. You are buying the 2028-29 steady state, roughly $8B a year, if every delivery lands. Watch the Anthropic financing, Monarch, and anchor prepayments. Underwrite the credit, not the narrative.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

What is real: real MW in low-cost power regions locked before the AI land rush, the Microsoft relationship, and $137M of balance sheet debt against $6.5B of customer prepayments. Everything else is a 2028 binary: Monarch 2GW on schedule, or the book starts to slip.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

Jensen Huang said in September that Nscale would have ~300k GPUs online by end 2026. A year later, with the Microsoft and Anthropic deals signed, the filing reports 25k active. The number selling the IPO is the supplier's, not the registrant's.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

The candid bits are rare in an S-1. 2024 revenue came from hosting crypto miners. The biggest 2025 customer, 73% of the year, is not named. Auditors flagged going concern doubt, resolved by a plan to defer or cancel capex. Same document sells 'exceptional delivery certainty'.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

The moat they pitch is power: ~70% cheaper than major US markets, owned sites in Norway, Portugal, Iceland and West Virginia. Fair. But GPUs are the big cost line at the same (rising) price as everyone else, and the revenue live today sits in colocation shells in Sines.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

So ~85% of the backlog is two counterparties, one of which is a direct competitor in Azure. The rest includes Figure, a robotics company with ~$19B raised committing $3.5B for up to 100k GPUs from late 2027. This is a credit book being sold as a revenue book.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

The book leans on two anchors. A $44.6B Anthropic deal, signed three weeks before the filing, runs off a West Virginia gas microgrid campus whose first 2GW is not online before H1 2028. The filing admits no binding financing commitments. Microsoft is the other ~42%.

000

George Z Lin @gzlin.bsky.social · 22/09/2026

Nscale filed for a NYSE IPO with a $103B contract backlog and $140M of half-year revenue. Only $2.6B of that backlog is active. The rest is campus that does not exist yet and financing that is not committed yet. The filing is worth a read.

000

George Z Lin @gzlin.bsky.social · 16/09/2026

DeepMind paper Dream-RSI converts logged discovery traces into frozen replay simulators enabling cheap off-policy dreaming to efficiently iteratively evolve executable exploration policies, foundational step towards self-improvements

010

George Z Lin @gzlin.bsky.social · 09/09/2026

9/9 The lesson is that span of control runs through the human over people, the human over agents, and the agent over agents. The winners keep every span small and every handoff clean, not one flat chart with a magic middle.

100

George Z Lin @gzlin.bsky.social · 09/09/2026

8/9 The people who actually run large agent stacks do not let one person or one coordinator watch everything. They group the work into small clusters with clean handoffs and pass only what the next level needs.

000

George Z Lin @gzlin.bsky.social · 09/09/2026

7/9 And its span is tighter than a human's. People compress context with trust, norms, and memory. An agent has none of those shortcuts, so every fact has to live inside a finite, expensive window.

000