As opposed to Benedict Evans yesterday on @LennysPodcast:
Benedict: "My most controversial opinion is that I think that Al is as big a deal as the internet or mobile and only as big a deal as the internet or mobile."
25 karma · joined May 16, 2015
Founding President & Operator of BlockchainU (2014-2016 --> rip)
[ my public key: https://keybase.io/schwentker; my proof: https://keybase.io/schwentker/sigs/56HCug_FgNVfs6HyA4ltZcYLSJ5w5rr5tSiiDSoqqfQ ]
Benedict: "My most controversial opinion is that I think that Al is as big a deal as the internet or mobile and only as big a deal as the internet or mobile."
Graber secured legal independence from Twitter before the Musk acquisition, positioning the AT Protocol as substrate rather than product. 30-person team, no HQ, "high agency, low ego" structure mirrors the decentralized architecture. She chose pragmatic adoption over purist decentralization when Dorsey pushed for purity, held ground, survived the $13M contract termination.
II. Reflection
Background spans digital rights activism, blockchain, crypto mining, privacy tech. The pattern: operator-educator who builds then teaches. Her AI framing cuts through hype: question isn't whether AI is good/bad but who controls it. Envisions users running own AI agents locally versus platform-optimized systems extracting value.
III. Question
Can protocol-as-substrate outlast platform incentives when VC returns require concentration not dispersion? Does pragmatic path toward mass adoption ultimately recreate centralization under different aesthetics? Can user-controlled AI agents compete with centralized providers given training costs and compute requirements?
IV. Paradox
Company must be expendable for protocol to become foundational. She explicitly unbundles company success from protocol success: "If the protocol becomes widely adopted, that's huge success" even if Bluesky fails. The accessibility that drives adoption creates lock-in that resists the promised portability.
V. Prediction
Near-term: 100M users, limited protocol adoption outside Bluesky. Critical juncture at 3-5 years: either multi-app ecosystem emerges or network effects concentrate around single implementation. Regulatory interoperability mandates (EU DMA expansion) could force protocol standardization. Wildcard: Bluesky fails but protocol outlasts company, another AT Protocol app achieves scale.
November 25 release, with RC on Nov 11—important that they’ve built in a 14-day validation window.
Working Groups & Interest Groups formalized, governance distributed.
MCP Registry now live as discovery backbone.
Async ops (SEP-1391) broaden use cases beyond quick tasks.
Statelessness & scalability addressed by Transport WG.
Server identity via .well-known makes discovery more intuitive.
Most popular protocol extensions being codified.
Tiering system for SDKs adds transparency on compliance & support.
Curious tension: MCP is positioning as both a protocol and an ecosystem backbone. Worth questioning whether tiered SDK support accelerates adoption or risks fragmentation.
Emergence suggests MCP is moving from “experimental” into “production AI plumbing.” If async, registry, & identity land cleanly, integration patterns could stabilize quickly.
1. Capability gating - Don't declare sampling capability during init for external/untrusted servers. Keep it enabled only for internal trusted ones.
2. Human approval loops - Force manual review before any sampling request hits your LLM. Protocol says "SHOULD" not "MUST" so implementation varies.
3. Token rate limiting - Set max_tokens params client-side when calling LLM APIs. Again, relies on individual devs following policy.
4. True MCP proxy - Terminate & reestablish connections (not just network filtering). Enables granular controls like "sampling for tool A but not B."
The real issue: first 3 strategies depend on individual developers following security policies. Only #4 gives centralized control.
Sampling's a double-edged sword - shifts LLM costs from server to client (good for internal workflows) but opens denial-of-wallet attacks from malicious external servers.
Most orgs probably don't even know this feature exists yet. Worth noting the travel booking example is compelling - instead of travel team paying tokens to format JSON responses, the requesting department's LLM budget handles it. Smart cost allocation if you can secure it properly.
Google just announced Agent Payments Protocol (AP2): standard meant to let AI agents securely transact across cards, bank transfers & even stablecoins. It uses cryptographically-signed “mandates” to prove intent & create an auditable trail from request → cart → payment.
Apparently 60+ partners on board (Amex, PayPal, Mastercard, Coinbase, etc.).
Google frames this as the foundation of “agentic commerce.”
Is it possible that w/i 6 months an agent-to-agent transaction for a large sum will go wrong, exposing the accountability gaps AP2 is meant to solve?
Is AP2 the missing trust layer for autonomous commerce, or just the start of a messy collision btw agents, payments standards & liability?
https://lfaidata.foundation/communityblog/2025/08/29/acp-joi...
reflection shift = coding less, defining problems & orchestrating more, like managing a fast but literal junior dev
question how do teams redesign review & governance so 3x output ≠ 3x risk?
paradox more automation → more need for human foresight, context & constraint-setting
prediction within 2y, top eng teams treat prompt-driven dev as default & code review as the new bottleneck
The final level represents multiple autonomous agents collaborating on projects, each specialising in different aspects of software delivery. This level represents a fundamental rethinking of software development, with agentic AI teams delivering alongside human teams. As above, given current tech, it’s not possible to achieve this level without human supervision and direction, so we consider it experimental.
observing how gen ai enters banking feels like watching a new instrument take shape:
fraud detection & personalisation sharpen yet real gain lies in workforce transformation
question
how do we cultivate human agent synergy not just tool adoption?
paradox
scaling automation increases need for human foresight
prediction
w/i a year ai agents run fraud detection 24/7 while employees focus on strategy, oversight & customer trust.
Windsurf Paradox: When Acquisition Looks Like Extraction
https://www.linkedin.com/pulse/windsurf-paradox-when-acquisi...
Windsurf Paradox: When Acquisition Looks Like Extraction
https://www.linkedin.com/pulse/windsurf-paradox-when-acquisi...
RAQ raq.com
Stanford University’s artificial intelligence leader Fei-Fei Li has quietly built a billion-dollar start-up in just four months, joining the fierce race across the tech industry to commercialise the technology.
Li, a computer scientist who has been dubbed the “godmother of AI”, created a company called World Labs in April, according to three people with knowledge of the move.
Incumbent saas players do not have legacy workflow dominance in that vertical
An AI-native workflow solves the most critical creative use case (collapsing the cost of creation and gaining right to workflow away from incumbents)
AI acts as a co-pilot empowering the user to coordinate across multiple existing Saas workflows. This is a particularly interesting category and could help create the next big super-app, especially in B2B.
However, the majority of AI startups today fall outside these 3 opportunity areas.
created by team of five women colleagues, Hyperledger education team & myself.
Success: A successful hackathon is where interests are aligned and everyone gets value from the event: the producers, sponsors, api providers, organizers, hackers, volunteers, event scribes, mentors, emcees, judges, audience, bloggers et al. Generally many things generally need to come together to make this magic.
Objective: What is the hackathon's objective? to test an api or tech (sf chatter 2010)? popularize it? build things on it (google wave hack '09)? cross pollinate apis? supercharge a conference (launch '15)? introduce a new tech paradigm (uc berkeley btc hack '14) start a developer program? build a dev program? find great devs or a great team (lady gaga backplane sxsw '12)? to find out whether to release tech (twitter annotations '10)? to pre-accel teams (angelhacks)? to have fun [go hack w/ rbodo @ singularityU:]? create solutions to problems (amex creative-currency '11)?
Prizes: can inspire (pp's 100k prizes), be complicated (sfdc $1m prize(s) '13 hack), be too little (as some have commented) or be just right - which depends on the objectives & aligning interests.
Setup: a page with links to all tech offered / available for the event. If setup is complex & lengthy consider creating virtual machines for devs (transparency & blockchain hack '14). Consider having an harassment policy (tcd sf '13).
Rehearse: if it's your first and you've many people coming, rehearse when doors open & devs pile in. Rehearse the hacks, make sure presenters can easily connect to the projector, have screen settings correct, sound on or off, have right sound level, know how to use a mic. If you've lots of demos - ideally have two podiums. If someone's preso can't start or gets stuck, you can quickly switch over to other podium.
Scribe: Have a hackathon scribe. Blog before/after the event. Create a tweet handle. During rehearsals create tweet for each pitch describing what it does & handles or name of at least the team leader. During each pitch, send out their tweet. Tweet each and every winning team. Tweet thanks to your sponsors.
Guides: hackathon.guide hackdaymanifesto.com socrata.com/open-data-field-guide/how-to-run-a-hackathon