Engineering brief
Agentic payments are near zero because API keys are the real competition
At a glance
The brief is enough
- Relevance
- Practical value
- Radar
- Time horizon
- This year
- Warnings
- Vendor pitch
X42 processed about $1M in 30 days and Stripe's MPP about $1.5K, while Apify paid creators $2M in a month. Agentic payment protocols have near-zero volume; API keys remain the practical blocker.
Agentic payment standards are near-zero in practice; API keys still win, so platform bets on agent payments are premature.
Watch if you are building an agent marketplace, selling an API service, or planning to give agents purchasing autonomy.
Summary
Near-zero volume. Despite a wave of agentic payment protocols from Coinbase, Stripe, Mastercard, Visa, Google and OpenAI, actual adoption is close to nothing. Curn, Apify's CEO, cites X42 processing about $1m in 30 days and Stripe's MPP about $1.5k. For comparison, Apify's marketplace paid creators nearly $2m in a month. The gap between announced standards and real transactions is the story.
The real competitor. Agent payments are not really competing with Visa. API keys are the incumbent default. Teams already give agents fixed accounts, sign up once, and pay by card through existing connectors. That is more convenient than setting up a crypto wallet or a new payment protocol. The need to let agents autonomously discover and buy new services only emerges when models are trusted with longer tasks and budgets, and when cheap token subsidies stop making it cheaper to build than buy.
Setup and trust. The second and third friction points are onboarding and discovery. Crypto wallets are opaque, and there is no equivalent of a trusted middleman for agent services. The speaker points to directories like Anthropic's connector list or Apify's marketplace as early trust layers. Without that, an agent has no reason to transact with a random API at an unknown URL.
Billing gaps. Early protocols only supported fixed price per API call. That works for simple data lookups, not for compute-heavy jobs or agent runs. X42's Exact scheme handled fixed price, but its UpTo scheme for variable cost let buyers spend money twice before settlement. Apify worked around that by charging upfront and refunding. Batch settlement, added recently, uses escrow to handle variable spend. The workaround shows how immature the tooling still is.
What's missing. The evidence is anecdotal and forward-looking. The speaker cites a forecast of a $15 trillion agent economy by 2028, and argues crypto will win because it is cheap and has no chargebacks. Neither claim is proven. Zero volumes could also mean there is simply no demand yet. The "agents will need budgets" assumption is plausible, but the timeline is speculation.
The signal. For engineering leaders, the takeaway is that agentic commerce is not ready for production workflows. Watch the funnel, not the protocol announcements. The unlock will come when large agent platforms integrate payments out of the box and when variable billing is solid. Until then, APIs with keys remain the pragmatic default.
My take
I buy the funnel argument. The volume numbers are too small to ignore, and the point that API keys are the practical default matches what I see in enterprise agent work. But the speaker has a stake in normalizing agent payments, so I'd be careful with the crypto conclusion. Low fees and no chargebacks matter, but they don't matter if the buyer still has to set up a wallet and trust an unknown seller. The bigger question is whether teams even want autonomous purchasing yet. Most don't. They want agents to complete tasks inside fixed, approved boundaries. That means the unlock is trust, variable billing, and a reason to let agents spend money at all.
What to do with this
- Ask your platform team to record how many agent tasks run through fixed API accounts versus any autonomous purchasing; expect the latter to be negligible.
- If you are considering an agent payment protocol, require a working variable-billing path with escrow or settlement before any pilot.
- Review whether your agent workloads need autonomous purchasing at all before allocating budget to payment integration work.
Watch the video
This video is blocked due to your privacy settings. To watch this video, please accept YouTube marketing cookies.
Related breakdowns
AI agents need identity-first security, not better sandboxes
Six real AI agent attacks and what would have stopped them. The surprising priority is identity management, not sandboxing.
Cheap models beat frontier models at writing like a human
Tone-Bench tests whether AI models can write like one specific human. Its most practical lesson: cheaper models with long prompts may beat frontier models on…
Why a faster model can make your users wait 8x longer
MiniMax's M3.1 flash is faster per token, but doubling concurrency lifted p95 wait from 1.5s to 11.3s. Evaluate the whole job, not the model.
Get TL;DW
Too Long; Didn't Watch.
A concise breakdowns of the AI and devtools videos that actually matter for engineering leaders.
Free. Weekly. No hype.
Video and thumbnails remain the property of their respective creators. tldw.news provides editorial analysis, commentary, and discovery links to original content.