Auth for Agents: Let AI Sign Up for Your Product with auth.md — Michael Grinich, WorkOS

AI Engineer · 21 min · 292 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:12Hello everyone.
00:13Good morning.
00:14My name is Michael Greenwich.
00:15I am the founder of work OS and I'm here to talk with you today about agents and specifically about how we can make agents more autonomous and allow them to access more services on the web.
00:29So let's jump in and get started.
00:30Shout out to those of you in the front row also.
00:32I see you out here.
00:35So about a year ago, software engineers were writing code like this.
00:39We were using AI, but we would prompt it and then we would write a little bit of code and then prompt it again and it would write some more code.
00:47And you would continue to do this human in the loop interactive software engineering.
00:52Later in the year, we had things like Ralph loops, if any of you remember those, where we kind of automated this.
00:57More or less this is the way that we were working.
00:59It was this human agent human agent back and forth.
01:03However, as the models got better and better, they were able to write more and more code and run for longer periods of time.
01:09And so today, a lot of software engineering looks like this.
01:12You write a single prompt and then the agent can spit out a lot of code.
01:15And sometimes it can actually build a whole feature, maybe even a whole product and maybe run for not minutes but uh hours and hours or even days and days.
01:25This has really transformed the way that we all write code.
01:28I don't write code by hand anymore.
01:30Probably you don't as well.
01:32And it's not just this one thread.
01:34You can actually parallelize this.
01:36So there's many different tools out there including, you know, the major code coding harnesses from the labs that allow you to run many of these in parallel.
01:43The limiting factor here, really the bottleneck is your brain.
01:46How much context do you can keep in your mind to keep all of these going at once?
01:50And this has transformed software engineering.
01:52People can be more productive.
01:53I think we've all seen in the industry how much faster things are moving.
01:58At the root of all of this is actually agentic engineering agents going from these token prediction language models to reasoning and now longunning processes that can actually do things autonomously
02:10for us.
02:11And what I think we've seen in software engineering here around agentic workflows is going to come to a lot of different categories.
02:17It won't just be engineers that have this way of working.
02:20It'll be people in many different fields.
02:22Agents are the next big thing.
02:26I think a good reference for this, a way to think about it is actually what's been happening with vehicles.
02:31So, you know, this looks like a pretty modern car.
02:33It has a, you know, digital display on it.
02:36This is probably what you would have bought in the last, you know, 5 to 10 years.
02:39But if you live in San Francisco or one of the cities with these autonomous vehicles, you've probably seen that we're starting to delete the interface.
02:46No longer do you get inside of the car and actually, you know, pull up your phone and get maps where you as a human are in the loop.
02:52There might be cruise control, but you're still in the loop.
02:54Instead, these are autonomous systems.
02:57You just say where you want to go, you get in the car, you don't even say anything, and it just takes off and takes you to your destination.
03:06I remember the first time I got into a Whimo, I was blown away, mind totally blown.
03:10And then after about 3 seconds, I pulled out my phone and started scrolling X. It became common place.
03:15And I think we've seen this already happen with engineering.
03:17The power of agents allows us actually to work and think at a higher level for us to, you know, exercise more of our executive function versus just planning and execution.
03:29There's a question here of what do agents need to be successful?
03:32Actually, when they execute, what do they need?
03:34Well, first they need a runtime, a sandbox.
03:37They need somewhere that they can run that needs to be safe, secure, performant.
03:41They also need tools.
03:43They have to be able to go do stuff in the world.
03:45An agent's not very useful if it can't take actions.
03:47So, it needs tools.
03:49Third, it needs context.
03:51You know, these new models with the power of the intelligence in them, it's kind of like taking the smartest person you know and dropping them into a company or a project or a team where they don't know anything, they can't really get anything done.
04:00They need to have context, information about the system, information about the goals.
04:06They need feedback.
04:07So a way that they can actually run and validate that they can check that their work is correct and if it's not correct keep going.
04:13And last but not least of course they still need human review.
04:17This might be code re review.
04:19This might be other forms of evals.
04:20This might be forms of checking their work to make sure that these agents are sort of aligned with your goals.
04:27The LLM the intelligence engine this new technology we've created in the last last few years you can think of as like an engine.
04:34It's like a really high performance.
04:36You know, I think this is a a V8 engine.
04:39You know, puts out a lot of horsepower.
04:41It can convert fuel into, you know, output.
04:45But an engine by itself isn't very useful for a car.
04:48You need to drop this into the actual car itself.
04:51You need the chassis, you need the transmission, you need the drivetrain, you need the wheels.
04:54And only with all of that is it going to be effective to take you places.
04:58And this is what a lot of people refer to as the harness.
05:02Harness engineering is this new domain where if the models get better but you don't have a good harness you know it's not going to be very effective so as the models improve we change our harnesses we adapt them to it and those things that I mentioned earlier are sort of a form of a harness
05:15so agents execute within these harnesses to get stuff done and they they need all those elements to actually be effective and when they drive they drive really really fast
05:27so say you have one of these agents that's going off you prompt it to go build something for So, you know, it spins for a while.
05:33It says, "Okay, I'm going to need this feature and this feature and I'm going to need a database and I'm going to need, you know, this editing uh, you know, system to deploy and I'm going to need, you know, maybe this thing for image resizing or video transcoding or sending email or whatever."
05:45And your agent can actually go do all this research and build all these systems.
05:48And today, what's happening with agents is they're actually selecting vendors.
05:53You've probably seen the rise of some of these systems where you know like the CEOs will tweet about their growth graph and it really takes off and it's because agents are picking them.
06:01They're picking their systems for sending email or deploying code.
06:04Their systems for storing data today actually it might be more important to build for agents than to build for people.
06:11Agents are kind of this new consumer class.
06:13I'll talk about that in a bit.
06:16There's this uh kind of popular thing that people say in San Francisco Silicon Valley.
06:20Make something people want.
06:22This comes from Y Cominator, you know, the startup factory from Paul Graham.
06:26And I think it's time to maybe edit this a little bit.
06:30And going forward, we also need to think about making something that agents want.
06:34There will be more agents in the future than people, maybe even today.
06:39They'll be faster to spin up.
06:40They'll do things quicker.
06:41They'll be making decisions on our behalfs.
06:44And so, if you want to build for the next era of consumers, you should be building for agents as well as people.
06:51But there's a problem today.
06:53You can access the web.
06:55You can access mobile.
06:56But agents themselves can't really access systems.
07:00Your business probably isn't open for agents today.
07:03Even though there's a lot of them out there, they can spawn very quickly.
07:07The door is not open.
07:10And we think this is because there's a missing primitive.
07:13Agentic registration, agent registration.
07:16There's a lot of great stuff around agents connecting and bringing context to systems and different open source frameworks, but that first step of an agent actually signing up for a service is actually missing.
07:27It hasn't been solved and it's a huge blocker to getting adoption.
07:33For the last like 20 years, maybe 30 years, automated traffic on the web has been bad.
07:40We've tried to block all of it.
07:42And so there's all these systems that we've put in place to detect automated actors and stop them in their tracks.
07:48DOS prevention, credential stuffing, and attacks.
07:51The login box, for example, is very hardened against agentic registration today.
07:56Not intentionally, but just because of the legacy systems of what we built.
08:01And it's very hard to distinguish between what's a good automated user versus a bad automated user.
08:07If any of you have used any of these like cloud-based uh browser execution environments, one of the selling features they often have is to break captures, which is pretty wild.
08:17You know, we're trying to go backwards.
08:19The capture today though is is still kind of dead.
08:22There's rampant abuse.
08:23There's lots of token fraud.
08:25And so, we need to find a better way to allow the good agents in and then block the bad agents, right?
08:30Because like I said earlier, if you don't build for the agent economy, your product might not win.
08:34It might not be successful.
08:38Today's signup flows assume that a human can read the landing page, that they can fill out a form, you know, put their email address, their password in it, that they can solve a capture,
08:47that they can verify their email, not just put it in, but go click something somewhere.
08:50Remember, an agent probably doesn't have an email address.
08:55If it's signing up for a service, it needs to have a, you know, a human can choose a plan to pay for it.
09:00You know, human can copy the API key out, paste it somewhere, use it somewhere else, and really even a dashboard.
09:05Think of a dashboard experience.
09:07It's not really agent native.
09:08All this stuff is built for people.
09:11It's kind of a human interface, not an agent interface.
09:15Agents need something else.
09:17They need native discoverability.
09:19Can they register for a system?
09:20Are the doors open for agents or not?
09:22Capability declaration.
09:24What can it do?
09:25Registration intent.
09:27Why are you registering?
09:28What are you trying to do within this?
09:29Or the intent to actually just sign up.
09:31There's a lot of stuff around risk here.
09:33So for an agent native registration, maybe you want to you're not sure if the agent signing up is going to be good or bad.
09:38So you need to actually do a risk assessment of that.
09:42There might be different forms of identity verification for an agent.
09:45If something like claude is going and signing up, maybe it can bring the user's identity with it or if you have a you know open claw or like your pi harness or something else, maybe it doesn't bring the user's identity and gets verified
09:57out of band.
09:59There's the whole thing around organizations doing this.
10:01So if I'm building something within a company, how does my company identity come in?
10:06You know, my organization, entitlements, permissions, credential issuance, auditability, the list goes on.
10:13There's so many things that need to be changed.
10:14In fact, probably most things need to be changed going from, you know, human registration to agent registration, agent native systems.
10:26I'm sorry to say MCP is not enough.
10:29Probably more than anybody, I'm a big MCP fan.
10:32We've done a bunch of events in San Francisco around MCP.
10:35I see somebody wearing one of our run MCP shirts up here that we made.
10:39Love MCP.
10:40MCP is great for connecting, for providing tools, skills, context, but it doesn't solve the registration step.
10:47Today, authentication through MCP requires your human to still be in the loop.
10:51You give consent.
10:52You add an MCP server, say for Post Hog to Claude, you sign in with your Post Hog credentials.
10:57What we're talking about here for agent native registration needs to happen without the human involved or at least not involved initially.
11:06So we've been working on this for a while and our proposal.
11:09This is a new spec we have called o.md because markdown files are the future right?
11:18Well, how does this actually work?
11:19What is o.md?
11:21Well, the idea here is that OMD tells agents how they can become legitimate users.
11:27It's a set of instructions that tells an agent registering for a service, this is what you need to do to sign up and be considered legit and for me to give you access.
11:36And the goal here is to give these service providers, you know, people that are building building services an agent native signup experience, but it's built on existing standards instead of inventing a crazy new religion or something really complicated.
11:50With OMD, we're not solving the whole problem around agent identity, you know, permissions, long duration, you know, connective services.
12:01We're just trying to solve this narrow one around registration because I think it's a huge blocker.
12:05Let's start small and grow from there.
12:07This is built on open standards.
12:08It's built on a bunch of work that's already been done across the OS specifically tailored to that registration step.
12:17often do you can answer what does the service do tell agents how to register it can tell them what identity proofs are accepted how to how how an agent or user is proving their identity
12:28what offflows are supported can you sign up with SSO or email address or MFA what scopes and entitlements can exist kind of what capabilities are here what free tier constraints apply
12:41you might have a system where you want to give an agent a little bit of free capacity but not too much until they verify or pay in some Okay.
12:47And then of course, how does the human get involved later?
12:50A human or an organization claiming it afterwards.
12:53So an agent can sign up, but you want the human to have custody later on.
12:58This is what A&D is designed to solve.
13:01These are the main questions around the registration step.
13:05So how does this actually work?
13:07Well, here's a little uh kind of cartoon demo that I'll show you.
13:10And all this works, by the way, you can go try it after this.
13:13So say I'm here in u whatever coding harness or system I'm using pi or clot or something and it says welcome back Michael at work OS that's actually my email if you want to email me do you want to keep working on your spelled app like yeah sure
13:25I'd like to share this with my friends but they can't access it and the agent's like oh well local host only works on your machine so your friends won't be able to hit it of course
13:34to share it I need to deploy it somewhere real some providers actually let me handle the sign up so you know you wouldn't have to sign up and do this you want me to look
13:45Yes, please agent look for services that advertise through author here does have an OMD and so the agent here can say ah Cloudflare that's the one cloudflare is a winner because I can go sign up for it
14:05I can access the service I can actually use it so I can do the whole thing you want to go with it yeah Let's do it.
14:14This is the registration step.
14:15This is kind of where the magic happens.
14:18What my agent is actually doing is minting an identity assertion and giving that to Cloudflare.
14:24And then Cloudflare is able to verify that or not.
14:27Cloudflare can choose actually to verify out of band.
14:30AMD is very flexible.
14:31There's different ways you can dial it in depending on the behavior that you're looking for.
14:35One size does not fit all.
14:37Every application is different.
14:38Constraints are totally different.
14:40If you're building something that's like a database service where there's going to be very little amount of information, you might want to give away a little bit of traffic for free.
14:47But if you're building something maybe like an email service or something certainly if you access the physical world, you you maybe don't want to give anything away for free.
14:55It's very application specific.
14:56So here Cloudflare says, "Okay, we're willing to let you deploy maybe for 72 hours or something like that.
15:01Get the identity assertion.
15:02Boom, boom, boom.
15:03Go through.
15:04It's called ID Jag is what the standard is."
15:07And then it returns back and says you're set up with a Cloudflare account.
15:10Now in this experience, my you know human I didn't go click anywhere.
15:15I didn't sign up anywhere.
15:15It just it just got the token.
15:18I can make a couple small tweaks.
15:19It knows how to use Wrangler.
15:20Want me to make those changes?
15:21Go ahead.
15:23Does anybody prompt their agents like this?
15:25It's just like yes, yes, do it, please.
15:27Not really saying much.
15:28It's able to write the Wrangler, get the thing set up, go through the full deployment process, and then boom, it's live.
15:36We actually have this working as a live demo.
15:38We did some collaboration with Cloudflare on this too.
15:40It's pretty cool.
15:41And you can tell the prompting that I was giving through this not exactly rocket science.
15:46The agent could just run through the whole thing.
15:47Actually, there's nothing actually that I was doing through prompting it other than just approving, approving, approving.
15:53It should be able to do this one shot.
15:57This is what it looks like graphically.
15:59So agent registration with OMD, you have your agent harness pi for example.
16:04within that is your LLM, maybe your memory, different, you know, context layers.
16:09It connects to your backing identity service.
16:11So this is the user identity.
16:13And when that agent goes and registers, it sends the discovery call to Cloudflare.
16:18It looks for that aund or whatever service.
16:20We also have this working with fire call.
16:21It's pretty cool.
16:23It sends that ID Jag, the identity JSON access grant.
16:28It's essentially a signed credential that represents the user says this is an identity receives back the access token actual API token and then boom it's it's done you can make normal API requests you can call MCP
16:42everything else still works so you can see that OMD is just for the registration step your normal API keeps working the rest of your docs don't have to change it's just that registration step for agents
16:54but with the simplicity comes actually an enormous amount of power because suddenly it lets agents do things end to end.
17:01I hate the planning step.
17:04When I write something, I just want the agent to go do it.
17:06Usually it kind of knows what I'm trying to get at and I want to come back to actually a running prototype and not like, hey, here's my long plan.
17:13Here's a bunch of services to sign up for.
17:15If I'm building new prototypes, especially, I would prefer it just to get deployed on something for free so I can see it and click around and not put me through the burden of going sign up for all these different vendors.
17:25And that's what OMD allows for.
17:27And if you're a service provider, if you're building APIs or services, this allows you to become a customer of those systems and you can think about it increasing your growth funnel
17:35starting from, you know, maybe just humans and adding the whole world of agents on top.
17:41Pretty great.
17:43We think this is going to be huge.
17:46Your door will be open for agents to actually sign up, register, use your services.
17:52This should be an immediate growth boom for your products.
17:54And I'm not saying everything that agents are going to build will be useful.
17:58Maybe not everything they build will go into production or last that long, but some things will.
18:02Just like with PLG or Premium before that, this is the way to grow your business and get more customers and actually become, you know, larger and more successful.
18:10And where there's usage, we believe there's also intent to pay.
18:15There's a lot of interesting work being done for agents to actually pay for stuff.
18:20There's crypto-based protocols like X42.
18:22There's other things that the payment providers are building, but we think that should be downstream.
18:26We don't want payments to be upfront.
18:27We don't want to have to force your agent to sign up with a credit card.
18:30That seems backwards.
18:31We moved away from that in the world of SAS.
18:34With OMD, agents can sign up, they can register, they can start using it and then downstream choose to pay on your terms based on your own pricing, based on your own model.
18:45If you don't believe me or on the headless type of products, this is Mark Beni off.
18:49He's the CEO and founder of Salesforce, the guys at the big building here in town.
18:54They believe in agents more than anything.
18:55Our API is the UI.
18:57Entire Salesforce and agent force Slack platforms are now exposed as API, MCP, and CLI.
19:03Talk about going agent fork first.
19:04A company that many people consider to be sort of a legacy SAS vendor really invented the category of software as a service.
19:11They're getting rid of their dashboard and UI.
19:14I mean, they're not getting rid of it, but they believe that agents are the future.
19:18So, if you're not thinking about this, I definitely encourage you to start building some prototypes, talking to users, because you don't want to miss miss this.
19:26If you want to use this, this is available today.
19:29We have open spec for it.
19:30There's a GitHub repo.
19:31You can actually have your agent go implement it.
19:32I know it's kind of meta.
19:34It's an open spec.
19:35Work OS doesn't control it.
19:37We have a version of it that we host, but you can build it yourself.
19:40You can run it yourself in your existing systems.
19:42I think this is so important that no single vendor owns this.
19:46It's one click enabled if you use our stuff, but you can build it on whatever platform you want instead.
19:52Last thing I'll close with, this actually happened almost 20 years ago, I think actually here at Moscone, which is pretty awesome.
20:00in uh January of 20 uh 2007 I believe Steve Jobs introduces iPhone and I was in college and I remember or yeah like first year of college I remember this super super clearly.
20:11It was the next great software platform that got built.
20:15So many companies got started because they could build on top of these devices.
20:18smartphone revolution and there hasn't really been a moment like this since I think in the the era of technology at least for me the stuff we did around B2B SAS and cloud and all that is great but this was a totally new paradigm a new software platform
20:35and I think agents are this agents are the next big wave and so what I would say to all of you is go out and build with AMD we're hoping to unlock this for you unlock this for your business and for your growth but we want to hear from you here the places you get stuck and I
20:49think together we can build for this next exciting era of software and build for the agentic revolution.
20:55Thank you so much