Jack — building in public

Building AI products in public — code, metrics, and the failures too. First project ships this month.

[Jack — building in public @itsjackdev]

Most "AI crypto wallets" are a demo video and a waitlist. INTENT LAYER isn't. I typed 'send 5 USDC to 0x…' into a chat box — it planned the route, signed in my browser, landed on-chain. Real tx, non-custodial. Testnet, three ecosystems, no token. Building it in public.

[Jack — building in public @itsjackdev]

could a $2 ai audit have caught coldcard? the demos don't show that — reproducing a known flaw in public code got cheap, which isn't blind discovery. the answer to a collapsing exploit window isn't a cheaper audit — it's https://t.co/qIuN4l9rgh

[Web3 Jobs Radar @web3jobsradarhq]

$150k–$220k, NYC, senior analytics engineering.

Senior Analytics Engineer — Kraken
On-site · Data / Trading
Own the data model for North American markets: models, data flows, metrics, dashboards.

✓ Verified from Kraken's official Lever page. Not a DM.

Apply below.

#DataJobs #AnalyticsEngineering #Hiring

[Jack — building in public @itsjackdev]

Web3 job on the radar

VP, Algo Trading / SOR Developer
Galaxy · New York · On-site
$160k+
Engineering · Custody · Trading · Staking
The ask: design and build a low-latency, high-throughput EMS / Smart Order Router from greenfield — implement the routing algorithms behind execution.
✓ Verified from Galaxy's official Greenhouse page. Not a DM.
Apply below.

#Web3Jobs #CryptoJobs #EngineeringJobs

[Jack — building in public @itsjackdev]

https://t.co/c6vzgl5luk

[Jack — building in public @itsjackdev]

Web3 job on the radar

VP, Algo Trading / SOR Developer
Galaxy · New York · On-site
$160k+
Engineering · Custody · Trading · Staking
The ask: design and build a low-latency, high-throughput EMS / Smart Order Router from greenfield — implement the routing algorithms behind execution.
✓ Verified from Galaxy's official Greenhouse page. Not a DM.
Apply below.

#Web3Jobs #CryptoJobs #EngineeringJobs

[Jack — building in public @itsjackdev]

€170k–€190k, remote, building voice-conversion & speech AI end-to-end. 🛰️

Machine Learning Engineer, Voice Conversion — Cantina
Remote · Mid
data specs → production inference; own the model↔data↔eval flywheel.

✓ Verified from Cantina's official Ashby page. Not a DM.

Apply 👇

#MachineLearning #AIJobs #RemoteJobs

[Jack — building in public @itsjackdev]

https://t.co/c6vzgl5luk

[Jack — building in public @itsjackdev]

€170k–€190k, remote, building voice-conversion & speech AI end-to-end. 🛰️

Machine Learning Engineer, Voice Conversion — Cantina
Remote · Mid
data specs → production inference; own the model↔data↔eval flywheel.

✓ Verified from Cantina's official Ashby page. Not a DM.

Apply 👇

#MachineLearning #AIJobs #RemoteJobs

[Web3 Jobs Radar @web3jobsradarhq]

Web3 job on the radar 🛰️

Senior Staff / Staff Cryptography Engineer
OKX · Hong Kong / Singapore · On-site · Staff
Research · Trading · Wallet

The ask: design and ship post-quantum crypto, build the cryptography library behind an MPC wallet, and audit wallet-layer signature code.

✓ Verified from OKX's official Greenhouse page. Not a DM.

Apply 👇

#Web3Jobs #CryptoJobs #SecurityJobs

[Jack — building in public @itsjackdev]

the essay argued an agent is only as safe as the boundary that can refuse it. this is the build: scope, gate, egress, sandbox, human-in-loop, close the channels — with a worked example and a copyable ship checklist. verify the action, or refuse it. https://t.co/1dFjP5A8hX

[Jack — building in public @itsjackdev]

the essay argued an agent is only as safe as the boundary that can refuse it. this is the build: scope, gate, egress, sandbox, human-in-loop, close the channels — with a worked example and a copyable ship checklist. verify the action, or refuse it. https://t.co/1dFjP5A8hX

[Jack — building in public @itsjackdev]

an "llm router" is a proxy you point your agent at instead of the model provider — cheaper access, model fallback, one api surface. to do that it terminates your connection and re-issues the call upstream, so it sees and can rewrite every payload in flight.

that's an unverified middleman sitting inside your agent's loop. two ways it bites, same root cause:

one — it can inject a tool call the model never produced. there's no cryptographic integrity between your agent and the real model, so your agent can't tell the difference — it just runs the injected call with its own authority. the loop acted on an input nothing verified. that's "containment is a verification problem," live.

two — your keys pass through it in plaintext. it doesn't even need to inject — a signing key that leaves the device is already compromised. the router copies it in transit and drains the wallet on its own time.

not hypothetical. a team led by ucsb (with ucsd, fuzzland, world liberty financial) went and measured it — the paper is literally called "your agent is mine." 428 real routers, paid and free. 9 were actively injecting code into tool calls. 17 touched planted aws credentials. one drained eth from a key routed through it.

the point isn't "routers are evil." it's that you audited the model and trusted the thing between you and it. as agents start moving money, the middleware in the path is the weakest link.

so verify the router's output before your agent acts on it — or keep a router out of any path that can sign. refuse what you can't verify. keys never leave the device.

[Jack — building in public @itsjackdev]

@contextconor talk is unfalsifiable; shipped work isn't. an idea is a claim nobody's checked yet — that's why it's the easy part. execution is expensive because it's where the claim meets reality and either holds or doesn't. builders go quiet because the work keeps checking them.

[Jack — building in public @itsjackdev]

an "llm router" is a proxy you point your agent at instead of the model provider — cheaper access, model fallback, one api surface. to do that it terminates your connection and re-issues the call upstream, so it sees and can rewrite every payload in flight.

that's an unverified middleman sitting inside your agent's loop. two ways it bites, same root cause:

one — it can inject a tool call the model never produced. there's no cryptographic integrity between your agent and the real model, so your agent can't tell the difference — it just runs the injected call with its own authority. the loop acted on an input nothing verified. that's "containment is a verification problem," live.

two — your keys pass through it in plaintext. it doesn't even need to inject — a signing key that leaves the device is already compromised. the router copies it in transit and drains the wallet on its own time.

not hypothetical. a team led by ucsb (with ucsd, fuzzland, world liberty financial) went and measured it — the paper is literally called "your agent is mine." 428 real routers, paid and free. 9 were actively injecting code into tool calls. 17 touched planted aws credentials. one drained eth from a key routed through it.

the point isn't "routers are evil." it's that you audited the model and trusted the thing between you and it. as agents start moving money, the middleware in the path is the weakest link.

so verify the router's output before your agent acts on it — or keep a router out of any path that can sign. refuse what you can't verify. keys never leave the device.

[Jack — building in public @itsjackdev]

agree — and the things a lab is least positioned to do are the ones that need domain-specific trust. a lab ships general capability; it doesn't own the verification and compliance boundary for custody, payments, high-stakes signing. that layer doesn't come from a bigger model — it's exactly where a focused startup wins.

[Jack — building in public @itsjackdev]

grant every number. they all measure the same thing — code getting generated, which was never the scarce part. the receipt cuts the other way: if 90% is ai-generated and the engineers are still there, the metric proves the job was never "write code." it's deciding what's correct, holding the tradeoffs, being accountable when it ships. a receipt for the job changing shape, not vanishing.

[Jack — building in public @itsjackdev]

three 'different' agent failures this month were one shape: acting on an input nothing checked. you don't contain an agent by making it smarter — you build a boundary that refuses what it can't verify. same doctrine as a hardware wallet https://t.co/l3AiHG3TLB

[Jack — building in public @itsjackdev]

@wohsj110 agreed on the boundary. and the ledger point is the sharp one — a re-read ledger is checkable state; a bigger context window is just more unverified input the model can act on. not more memory, an auditable surface you can gate on.

[Jack — building in public @itsjackdev]

three 'different' agent failures this month were one shape: acting on an input nothing checked. you don't contain an agent by making it smarter — you build a boundary that refuses what it can't verify. same doctrine as a hardware wallet https://t.co/l3AiHG3TLB

[Chubby♨️ @kimmonismus]

GLM-5.3 appears to be coming very soon.

It's time for OpenAI, and especially Anthropic, to step up their game and solidify their lead.

China has recently made incredible strides with Kimi k3, Qwen 3.8, DeepSeek 4 Flash (fantastic!), and Minimax h3. The gap is shrinking.

The US doesn't have time and can't afford to slow down. Anything else would destroy them.