Spending is not intelligence. An agent that executes every purchase is not smart money. Judgment is knowing when not to spend.
Jim Chang (@jimchang), VP of Product, @lifiprotocol, on why the next layer of agentic commerce is not execution but decision making:
and btw, if you are considering building with intel tdx we have https://t.co/Fga2dQkmby
it lets you to:
- generate tdx attestations
- verify them
- bind public values as part of tdx 64 bytes of reportdata
- gcp and azure support
(similar taste to zkvms but with tdx guarantees)
Love seeing teams get hate when they move into TEEs
- team: we now use a tee
- zk/fhe/mpc people: tee(dot)fail, omg, ngmi
- team: yes
*proceeds to keep winning*
@trent_vanepps@andrewwilkinson@drahcir@peerxyz TEEs are arguably the only solution that has an actual good performance/verifiability/guarantees balance.
As long as you don't have physical access, you are safe any of the flavours (AWS Nitro/Intel TDX/AMD SEV).
If they want to scale to 1000s of users is the best approach imo
blockchain are verifiable by design and the human has always been the arbiter
we are now moving into a world where a machine, outside the chain, act as a human and thinks by itself
Ep.6 de nuestro especial "Háblanos de tu proyecto".
Aqui hablamos con @livylabs, una compañia que te permite convertir afirmaciones externas en datos en los que puedas confiar.
the first $1m agent mistake won't look like a hack.
it'll look like a valid transaction.
• the agent had the key.
• the api call worked.
• the wallet signed.
• the log was saved.
but the action was still wrong.
that's the problem nobody wants to say out loud:
agents are getting payment rails before they have accountability rails.
stripe, visa, mastercard, coinbase, cloudflare, everyone is building for agentic commerce.
but identity is not accountability.
> knowing which agent acted doesn't prove the action was allowed
> proving code ran doesn't prove the decision was right
> logging after execution doesn't stop the loss before it happens
the missing layer is a resolver before execution.
one question before value moves:
was this exact action allowed under this exact intent, evidence, and policy?
that's what we're launching at livy.
> agent proposes
> livy checks evidence
> livy applies policy
> decision returns: APPROVED / REJECTED
> execution only happens if the transaction matches the approval
> receipt links the final transaction back to the decision
starting with verifiable @solana actions.
• not another wallet.
• not another dashboard.
• not another “trust me bro”.
wallet policy controls the signer.
livy controls the action.
• agents propose.
• livy authorizes.
• execution follows.
if you're building agents that trade, pay with x402, use wallets, touch defi...
livy is the way