Why APIs Are Still Broken in 2025 - Agentic Dev with Speakeasy | Episode #101

Why APIs Are Still Broken in 2025 – Agentic Dev with Speakeasy | GTwGT #101 - YouTube
https://www.youtube.com/watch?v=BQzzVSoSnn4

Transcript:
(00:00) apis are the backbone of modern software yet building and maintaining them often feels like navigating a maze of outdated docs inconsistent SDKs and fragile integrations throw in a little bit of AI MCPs agents etc etc and it becomes even more cloudy speak Easy is changing that automating SDKs Terraform providers and documentation directly from your open API specs today I'm speaking with Simon Newu co-founder and chief operating officer of Speak Easy to explore how they're redefining API development this is episode 101 of Great Things with Great Tech with Speak Easy
(00:48) hey Simon welcome to the show episode 101 thanks for joining me and let's let's get into the world of AI tooling MCP agents and all the great stuff that people are talking about today a very timely conversation obviously uh but you know before that let's get into your background and talk about you know where you started in IT and how you became to co-found the company speak easy  well thank you so much for having me Anthony um great to be here congratulations on hitting uh master a number of episodes
(01:21) um yeah happy to dig into my background so I started my career actually in consulting uh as as a management consultant and then did some investing uh I was an investor at a private equity uh firm that actually invested in tech companies  okay  having having spent sort of a few years doing consulting and investing kind of got the operational itch you know it's it's uh consulting and investing obviously great jobs but just kind of feel a little bit removed from the day-to-day and really wanted to for myself see what it was like to build a company build a product with my own hands and so uh I ended up moving out to
(01:58) Silicon Valley uh obviously the place to be if you want to work in tech um and uh found a job out here and have been here ever since which is kind of shocking i was supposed to be here for about a year and uh it's been for 14 now so  there you go so yeah I mean obviously your accent there it's British it's it's it's it's quite proper it's it's great you haven't lost that for 14 years so yeah how how do you find that so you know like obviously you know doing what you did in England you had that itch to come and
(02:25) operationalize yourself and and actually start a company which is great i think every every person who's got that entrepreneurial spirit should try this one time in in their life to see what it is and go through the rigors of it but how did you find the early years i'm interested you know like was it everything that you thought it would be you know 14 years ago clearly something worked for you to be there for 14 years yeah i'm not sure if I'm here because of the weather or because I enjoy the the kind of work actually it's not entirely clear  the weather's easy right yeah it's not hard to beat London weather  exactly
(02:56) right um no but I mean um the earies were tough the ears were tough because you know it is such a huge uh sea change going from being an investor being a consultant where um it's just a whole different kind of work right and being operational um I started my career as a as a product manager at a company called VMware building product is just such a different uh mentality it's such a different style of working right consulting and and private equity it's much more project based it's much more outcome situated right and so you're really thinking how do Hey how do I do this for the next four weeks and I'm
(03:31) driving towards a particular end goal um product management can be like that you're trying to build a product and there's obviously milestones but it's much more longunning and you know just the cast of characters you have to work with are just you know completely different to build and launch a product versus you know investing in a company  can I can I go back a little bit so VMware so what product were you product managing at VMware because obviously my a lot of people listening would be very obviously very familiar with VMware and the products so what were you product managing this this takes me way back way
(03:59) back in the day this is 14 years ago um I was in the core vSphere product team and I was responsible for sort of a few vzero products um principally among those was a utility called um it's essentially a phone home telemetry kind of product where basically um one of the big um pain points with vSphere is understanding um your entire deployment of vSphere understanding what sort of issues are occurring in your virtual machines you'd have these massive corporations with you know thousands of VMs right um how are they provisioned what issues h are happening how do you make that support
(04:40) proactive rather than reactive it was sort of like a kind of a um I guess kind of a novel thing that people were doing back in the day the sort of idea of proactive support uh and so I sort of built the first version of that  wow okay yeah i mean it's right in the wheelhouse of when I was big into it right 20101 was when I was really kneede in platform engineering and trying to make sure that the vSphere platform was running and you know so I was I was there that's quite interesting right so you were you were there trying to help the the support which you know to be fair even 14 years
(05:12) later support is a big bug bear and a big issue with a lot of tech companies right because you just can't seem to get it right so anything that can help that is really interesting and I wonder whether AI and agents and whatnot can actually help that moving forward but we might we might push into that a little bit later on  right right yeah but so okay so VMware and then you know you've gotten into a few different companies after that so what was the sort of genesis of meeting up with Saga who's the co-founder of Speak Easy to get that going we're fast forwarding a little bit but let's let's get to that point  of
(05:43) course yeah definitely um the story of how I met my co-founder Sagar is a very typical uh Silicon Valley not typical but maybe a kind of story that can only exist in Silicon Valley I think actually um we were essentially introduced by a friend of mine who works in venture capital and SAR was speaking to him about an idea that he had you know squinting and sort of 30,000 idea of of a a pain point that he was thinking of working on and he was looking for a co-founder and um you know my my venture capital friend um said you should speak to Simon you know he's uh entrepreneurial he's always enjoyed
(06:23) starting companies you guys should talk together and We met at a coffee shop and sort of had a two-hour conversation about each other and the idea um spent a long time getting to know each other just working together you know for a few months  and then and then sort of ended up founding this company  yeah okay that's not that's not it is a is a very Silicon Valley sort of story in terms of you know you're in and around the gravity of venture capitalists and technologists and people that are just into that
(06:48) entrepreneurial spirit like we touched on early on um but it's not that sort of typical start where you know there's I mean I've spoken to a few of late where they're working inside a particular company they they've developed a tool that's internal and then they might pivot that tool out into a into a company right or they might donate it to Apache and then build some enterprise sort of solution around that that seems to be happening as well this is more I guess a crafted sort of coming together of the company right and someone's got
(07:17) the ideas and you've got the operations you've got the experience let's kind of make this thing happen so what what was Saga's idea to start with what's speak easy sort of um core founding principle yeah definitely i mean Saga was um at a company called Live Ramp which is um a large company um involved in sort of the advertising uh identity kind of space and he had basically built a framework for internally developing APIs and making them secure um evolving them publishing them externally right making
(07:53) that developer journey uh much much simpler the basic insight is that APIs are fundamental to how software communicates probably people don't realize how fundamental APIs are but even things like you know you open up an Uber app and you want to order a car right uber has so many services involved in understanding car availability booking that availability routting pricing that's it just internally and then externally for the APIs it's things like calling out the payment processes um but the but the key insight is that despite how fundamental and critical these APIs are API producers don't
(08:33) really have a lot of tools available to them to build great APIs API consumers struggle with integrating those APIs and so for this thing that's so critical and ubiquitous and important um the surprising little tooling i had noticed this because at a previous company I was um a product manager and one of the things that we had to build for our product was an API and I just couldn't believe like how hacky the whole thing was you know we're just like building this thing by hand and it was just like wait a second isn't every developer having this problem how how can this not
(09:02) be a solved problem um and so you know I think when I met Zo and we talked about that that's sort of where the idea came from  yeah yeah and I think you've hit the nail on the head because a lot of APIs if if I look at um in the world I'm in now you know we've gone through a bit of a rewrite of our APIs from an old sort of HMLbased format to modern JSON and open API standards but but still it's troublesome right and you still get a lot of our customers um you know not able to consume them right a new version or a vision comes in and it breaks a lot
(09:31) of other stuff um you know so there is historical issues with APIs as great and as fundamental as they are they're a pain in the ass for people to play with right and and to to use and integrate I remember back when I was um working on um the the platform as a hoster we would create the the manual flow of the product first hand it over to the developers and they would then automate it and do their thing but even working with those APIs back then it was it was just it was just a pain right and then a major version comes out breaks it So is
(10:02) this kind of the fundamental you know uh value prop and reason for being for speak easy and I think from the naming perspective speak easy I h it kind of makes sense but am I wrong or am I right there are we just speaking easy together in this world of APIs  you know it's so funny because we sort of redconed that description the true story is that uh Zod likes speak easys you know like the bars I do  I think he doesn't like a good speak easy um but we definitely redcon the idea that you know AP PI need to speak easily to together so
(10:35) I love the dual a dual a dual meaning i love it no I I mean it works either way doesn't it so yeah go to speak easy like the speak easy but yeah it's it's kind of makes sense and it fits what you guys are trying to do right um absolutely  so yes okay so in in the leadup how how do you validate that idea and what what actually is a company doing what what's your product set what's the core platform that you're offering  yeah I mean how do we validate the idea i mean I think the northstar has always been you know as we discussed right how do
(11:03) you make well our vision official company vision is how do you unlock the full potential of APIs right and that's always been the north star how do you make it easier for API producers service providers to create better services how do you make it easier for consumers uh either human developers or as we'll talk about I'm sure uh agent agent consumers how do you make it easier for them to to to consume your API um how we went about validating it is a much more typical Silicon Valley story Right we probably spent a few months building MVPs and iterating on different ideas and there
(11:34) was a few different iterations that we went to but basically towards the end of our first year or within the first year of our um existence we settled upon the uh what is sort of still our flagship product today which is um SDK generation right and it's really focusing on what I would say sort of the the API consumer journey um how do you make it as simple as possible for an API consumer to integrate with your API and to get start start using it you made the point that like API um APIs haven't really changed
(12:05) that much in the past 20 years and I think it's absolutely true the state-of-the-art for many companies today is still give an API consumer an API reference dock and that's it right if you think about Stripe 10 years ago people were like Stripe is amazing because they have great API documentation and that's true that's true but that's like such a low bar right and so um our our thesis is that what you really want is you want um you know rather than giving all of your API consumers an API reference dock and saying go build it yourselves and you
(12:36) know 50 consumers are going to build the integration 50 different ways and they're going to read the manual incorrectly what you really want to do is you want to give them a single code library an SDK a software development kit um that basically wraps the API and does the work for them they import the your consumers import the library um and then they just call the library and then job done as the API is integrated with yeah and that's that's it is a fantastic northstar again very Silicon Valley sort of term that we're hearing more lately in terms of what what the reason for
(13:04) being is but yeah and I've I've used it like we we talked earlier that I went and had a look at it because you know effectively hand you feed in the open API um schema put it in there and then it's got like a command line you've got the web UI and then that that quick start guide effectively sucks that in and then you've got the ability to create the SDK based on you know a couple of different options which is I think the code language you know so I saw I saw all sorts every kind of major code language that you can imagine for and Terraform which we'll touch on a bit
(13:33) later on as well um and then the idea is that you it does it interrogates it creates the SDK and then spits it out for you to then use in a much more simpler concise way that's kind of as I saw it right it's it's kind of simple but no one's that I've seen has done this before So I think this is where the the value of the company really should come to light  absolutely yeah that's where we started you know so absolutely the idea is to be able to take in an open API specification a description of your API and then output documentation
(14:03) output um SDKs um output as well Terraform providers as you mentioned and also recently MCP servers agentic tooling right and the idea is absolutely to make your API as easy to integrate with as possible it's worth noting that you know we talk about the journey uh the the journey we just talked about is really um an API producer journey right it's you building an API and you building an API can be a developer internally uh for another team somewhere else internally at a company and it can be externally as well right
(14:38) it could be um you know if you're Stripe for example and you want to offer great APIs externally it's that too right so how do you offer these internal or external consumers human developers or agentic consumers an amazing uh experience  yeah and I think that if we level set about I guess what an API really is I actually generally think that a lot of people probably don't know that because I know that until I started writing my own sort of APIs or actually help prompt engineering my way into writing APIs i won't claim to be a
(15:06) developer like I said right I'm a hack but prompt engineering has opened up the ideas in my head um and then actually seeing what makes an API as as a application programming interface i think a lot of people think it's a little bit of magic and quite complex but effectively an API is just is it's a it's a it's a function that's exposed to do a thing and then you can talk to that um through a front end or whatever it might be to actually you know speak easily together if I can use the pun so how would you explain what an API is to
(15:35) someone who asks you what are we doing here  well it's hard to do better than what you just said to be honest with you but I'll do my best i'll give an example of you know a company that I worked at Flexport right um yeah you know flexport there's all kinds of services internally sorry flexport is a um digital freight forwarder so what we do is um if you are a e-commerce store in the US you need to ship goods from China right how do you get containers from the port of Shanghai into the port of LAX and then how do you
(16:07) get it from LAX onto a truck to your to the destination warehouse um there were all kinds of services internally that we built at flexport that are involved in that loop there is for example a service that knows the sailing schedules of ocean ships around the world and there's a service that knows how to create orders from a customer and there's a service that knows how to process an invoice and like issue a payment request for a customer all of those services work together through as you said APIs it's how these services communicate so when um a customer presses submit on an order in
(16:42) the web application it goes to the right service that service communicates with the next service through an API and so we said at the beginning how important how critical APIs are it is critical because it's how software communicates it's how teams communicate internally and it's how uh you call third party providers as well  yeah they say software is aiding the world and AI is catching up but APIs you know are the main meal right they're doing everything at that point um that's that's that that is an interesting part and I think from from
(17:12) that perspective how uh how's speak easy you know creating that um I've lost my train of thought there but how has speak easy created the connection so how does it how's SDK come at the end of this particular process  absolutely yeah um so if you think about um a typical journey of an API consumer So we talked about um again let's take a example of a digital freight forwarder um different teams will create different services a team a team might create a service that exposes the sailing schedules of ships um the state-of-the-art today as we just talked
(17:57) about is I've created the API maybe here is an API reference doc and the API reference doc explains what the API does explains how to call it and it explains the kind of data that the API returns right in the case of the sailing schedules it would return some kind of object that says on a Monday at like you know a ship will arrive in the port of Shanghai and then on Tuesday it will be in the port of XY Z right  mhm  um that sort of like bare minimum sort of API right so where do SDKs come in well as I said like if you're just giving your
(18:30) consumers an API reference doc you're asking every person who wants to use that service to read the read the manual which no one ever does understand the manual understand the return types understand what the API does um they're all going to forget about things like um if the API doesn't respond for whatever reason right latency traffic you know the server's down for a few seconds they're going to forget about retries um your API consumer is going to forget how to write authentication correctly and so
(19:02) basically you're asking your API consumers again whether they're internal developers or whether they're external users of your service you're asking them to read a manual and think about all of these things and you're really putting up roadblocks  um which is kind of crazy because you want people to use a service while you built it right you want people to use the thing that you built  seamless and efficiently yeah  exactly yeah exactly right and so where do SDKs come in um SDKs are basically um rewriting
(19:28) the integration once and giving your consumers that pre-written code that says here is how you integrate with the API  get it yeah  i have taken care of authentication for you all the key variables that you need to know your API key or the oorthth you know token exchange and um the retries have already built in pageionation right when the endpoint returns many many results in different p pages of results i've taken care of that you've done all the pre-work so all that a developer needs to do is import the library and call the right method and so it's really
(19:59) dramatically cutting down on the integration  we work with a lot of customers all kinds of industries but I'll take fintech as an example fintech is a very complicated space um and often times if you're a bank or a payments provider the integration journey for an API with just an integration API doc might be weeks or months and you're jumping on phone calls with your API consumers trying to talk them through it well SDK has dramatically shortened that because again the end user the end developer just imports the SDK and then calls the rep method to go  yeah absolutely i think that the point that I
(20:33) had before which I've just remembered as well is that you know APIs generally in today's world will cost there's some cost associated with it whether it be overheads through the code or even the the virtual cost of consuming API a lot of a lot of companies charge for an API right it's become a bit of a business business um outlet how and I guess when you when you're actually being charged for something or you are charging for something you want it to be as efficient as possible i think that kind of ties
(20:59) into what you just talked about there as well the efficiency of making sure that you're calling these things in the right way so number one you don't blow your budget or number two you don't you know cause an overload or something in the system which actually renders it useless oh you're absolutely right i mean the way we think about the market here is there is a set of companies for whom the API is an important product surface stripe is of course the poster child and if if the API is an important product surface you want to make it as easy as
(21:29) possible to integrate with the longer someone can't integrate with your API that's just like time that you can't get revenue right and um again Stripe's a poster child  the reason why Stripe is so successful is because they made it really easy for developers to integrate with stripe were not the first payment processor by a long shot but they were the first to make it really easy to integrate with  um so absolutely that's true  yeah and I I I you know one of my little side projects is is was working
(21:54) on with the Stripe integration and trying to get that going and and yeah to be fair I didn't find it too difficult actually even in my you know lack of um you know opportunity to actually develop i was like "Yeah I can do this quite easily.
(22:10) " And here's a little function here's a call they've defined it just put it in there and it should work and and it did so that's it is a really good example of that  so what you're trying to do is you're trying you're trying to bring that sort of mentality and ease of use to the rest of the world  exactly exactly i mean  finding developers who want to focus on API development experience who want to build great artifacts like docs like SDKs for the API it's not easy right if you're Stripe great sure i hire a team you can hire a team for that but most companies 99.
(22:36) 9% of companies can't do that right and so how do you bring that experience to companies that are offering a public API for which the API developer experience is blocking time to revenue internally how do you bring that same ease of consumption to internal development teams so that you know if you're a large company with hundreds of thousands of developers how do you make it easy for the team in Boston to consume the service that the team in Singapore has created right it's the same it's the same problem they're facing the same consumption challenges um yeah one might
(23:07) not be revenue one might be more developer productivity but you know you still want to make it easy  yeah yeah and more and more we hear this that the Amazon story is built on the fact of departments trying to work together and this is what creates a product set that they've got today and the services that's that's kind of the classic Amazon Web Services story right and and APIs were core to being able to for different department to talk to each other um let's let's let's segue into the AI conversation agents but I want to start by saying so when how much of AI and you
(23:38) know generative AI and platforms and LLM are you guys leveraging in the back end to get this happening because I'm not I'm guessing there's not a group of guys and developers sitting in the background trying to sort of suck it in and kind of spit out the code that after you've inro the open API specs so what part of AI has enabled you guys to to make this a reality  such an interesting question um you know surprising many people when they first hear about speak easy and how fundamentally what you know one of the things that we're doing is we're taking an open API spec we're generating code
(24:12) SDKs from the open API spec one of the first things people assume is oh you must be using AI to do that right  um funnily enough we're actually not necessarily using AI just for that or specifically for that and the reason is that I think um everyone is kind of realizing There are certain things that AI is very very good at and certain things that AI is not very good at and it turns out that if you want to generate um this SDK code um which makes it very easy to which is you know idiomatic you know so such that a TypeScript SDK looks like TypeScript
(24:46) code and a Java SDK looks like Java code um if you want to make it idiomatic if you want to make it featurerich so that it has things like authentication and pageionation and retries built in it turns out that um AI can't necessarily be relied upon to give exactly the right output that you want every single time you know we the limitations of AI are kind of well known at this point in time you know some well known with hallucinations and so on  and so we basically have built um a deterministic generator actually  okay but AI is very
(25:19) good at things like given an open API specification um which again is a description of the API often times that open API description may not be enhanced it may not be very rich it may not have all the descriptions it may not have all the examples um the method names of the in the open API specification may not be um very uh descriptive and so AI is very good at looking at that and saying hey here are suggestions and so one of the areas that we use uh AI for is that other than open a uh SDK generation we
(25:51) are doing a lot more around like the open API editing experience the open API evolution experience and there AI is is um is heavily involved  um yeah and and so there are other sort of product features that we use AI for but but I guess the bottom answer is like you know very targeted sort of use case that that's right yeah it's so and that seems pretty smart because there are parts where it's good it makes sense there's other parts where you probably can't trust it so much to get the desired quality of the output that you want and what you guys want because ultimately speak easy is here to make people's lives a bit easier but if you're
(26:24) muddying the waters with a little bit of maybe hallucination or whatever it might be it's not going to work that way so I totally get that um so what's really interesting is when like I said we I kind of played with it over the past couple of days and you put in the open API spec and then it's like okay let's set up your workflow and then it goes what language you would like to generate and like you we've got I'm reading it here because I wanted to to get it right python Go Java C which is interesting Unity PHP Postman and Ruby so you got
(26:53) you got everything covered and Terraform obviously as well so when So just explain that process so you put in the spec and then it's like asking you for the language and then the SDK is created based on what you select effectively right  that's exactly right yeah i mean um that's sort of the SDK generation workflow where you give us a spec um actually once you give us a spec we are working on um an API open API editor experience where you'll be able to automatically enhance the specification as I mentioned once that is done once you're satisfied with the specification
(27:30) um you select the languages and then our deterministic codegen kicks in it generates idiomatic featurerich um SDKs in a number of languages um and there's also a publishing workflow where if you're generating a let's just say TypeScript SDK that gets published to package manager and that's just sort of what developers expect developers expect their SDKs their packages to be available in these package managers we really handle the full life cycle everything from how do you define an API how do you write about it how do you
(28:00) describe it in the open API spec all the way through to the consumer the end consumer how do I consume that API  yeah that's awesome so okay let's move on to MCP times like I I always say this but we could talk for hours on this and it's a great conversation um really hitting a lot of key points today for a lot of the listeners let's talk about AI generative AI and MCP specifically an agent to agent but let's talk about MCP because you have specifically gotten something in there um and like we talked about MCP has been around since November only
(28:32) November but feels like it's been around forever it's dominated all the socials and the trends and whatnot uh and everyone's even vehic you know productizing it in Speak Easy yeah um would it be helpful actually to talk a little bit about why MCP is important for your listeners and why that's  that's good yeah yep yep absolutely yeah well um I'd like to give the example of um well firstly sorry AI is obviously extremely powerful right everyone's used chat GPT and you know you can ask us some questions and and that's fantastic and it'll give you some amazing responses but chat GPT and
(29:10) Claude and all these other LLMs um they're giving you information that's based upon static data that they were trained on as of a certain date right and importantly they can't take actions on your behalf if we want AI to sort of take the next leap forward if we want AI to truly sort of um be agents on our behalf and to take actions on our behalf and to find out the information they have to be able to access APIs right i'll give the example of um let's say you want an AI agent to help you book a trip somewhere i I want to ask an agent to book me a trip for my next vacation well the agent's going to need the AI is
(29:43) going to need to know um the dates when I'm free so it needs to like plug into my Google calendar and figure out when I'm free it needs to query Expedia or Google flights it needs to understand what flights are available to which places it needs to understand um the ratings of hotels in the area right and then ultimately needs to like book something for me and all of those things require APIs and how does an agent talk to an API that problem is what MCP solves it's the thing that sits between an API and an agent and allows the agent to communicate in a language that it
(30:13) understands uh and to take these actions to query on on its behalf so NTP is incredibly important um and because speak easy generates uh we understand APIs very well we understand the open API specification we can generate documentation we can generate SDKs we can generate terraform providers we can also generate agent tooling MCP uh on behalf of our API customers that makes their API agent accessible  yeah and is that in the same way where you know you you create an MCP server for in Python Go Java all the different sort of um
(30:46) languages that exist at the moment or is it  That's exactly right  in fact if you on that on that list on that little CLI output you're looking at right now if you were to select TypeScript what you would find so we bundled our MCP server generation into the TypeScript SDK just for sort of um rapid release you will find in that task SDK there's a little source folder and if you click that folder and you click if you click that folder you'll find another folder called MTP server and that basically has for every single one of your API methods it will have a tool the language of MCP is
(31:16) tools and so that is  that is that is MTP server you can load that into claude today and you can start querying your APIs  yeah might might give a crack later on for sure um okay so MCPS and then more lately we've heard agent agent or a a2A so that's sort got up a bit so how is that because actually so last year we're talking and what you described in terms of LLMs and models being you know having parameters and it's kind of set and done and it doesn't know past a certain date where they wrap it up and then rag came into
(31:45) play and rag was a good way to talk to APIs but rag wasn't a way to I guess do what you just said which is the agent experience which is to go to multiple um different API sources and pull it in together or maybe you could have done it with rag but it was very much not based on any protocol effectively right so I guess where MCP comes into play and what I've heard is that it's effectively and how anthropic actually talk about it it's a USB standard for this world right which completely makes sense then you got agent to agent so maybe just explain the jump from MCP to agent to agent or
(32:20) the sideby side of those two as well just quickly  yeah I mean um you know I I firstly want to say that I am not like an expert in any of this I'm not sure who is an expert things are moving so so fast  that's right  yeah But I mean the general idea here is that like MCP opens up the opportunity for agents to talk to your APIs um and you know what comes next um we actually think that what comes next is building software um with agents and actually how do humans build software with agents firstly um so there's a whole like
(32:53) there's a whole mountain to climb around like what happens there and then of course agent to agent you know they they will need some way to communicate some protocols to to communicate and so I think there's a lot of um innovation happening here around MCP I'm not necessarily up to speed with all of it to be honest with you but um  but yeah I think MCP is going to be a really important way that I think there's also a lot of other agentic protocols that are coming up around how agents communicate together
(33:17) so  um  it seems seems like there might be a little bit of a war you know in terms of there's always when whenever this happens in life beta versus VHS you know that there's always this sort of protocol war that happens and I feel like you know MCP has taken the lion share very quickly it seems like it's going to continue and it's not going to just drop off the face so it's not it's not going to Betamax itself basically if I can say that um to sort of analogize it but yeah it's it's good that everyone is on to it because it does create this world you had me thinking though when
(33:48) you talked about that use case right about booking the trip and you know all the ratings and whatnot we have to get to a point where as a human you have to trust implicitly this agent right and I think that is going to be the biggest step like we're talking kind of futures now and whatnot but the time is is is for it so for me if I'm looking for if I'm going to trust an agent to go and say book me I'm going to so I'm going to Queenstown with the family um in New Zealand in in a month if I'd got the agent to book that today I'm not sure
(34:18) that I would be completely 100% okay with it right so  I feel like we're we're a while a while away from that but maybe we're not and maybe this is the the great trajectory that we're heading to where agent agent MCP is just going to allow us to go hey I want to book something in Queenstown you know my par you know exactly what I want what I'm looking for hotels Airbnbs my flights um I want to do um activity XY Z go make it happen that's that's kind of where we got to get the trust level to right  i I think you're absolutely right yeah i
(34:49) mean um I guess the point there is that there is so much that needs to happen around MCP to make agents agentic workflows useful um agentic workflows meaning like an agentic workflow that a human developer describes or you know agent to agent sort of workflows as well um MCP and its usage probably is going through a similar I hate to say sort of hype cycle as like AI in general right you know sort of exactly yeah a couple of years ago everyone was like AI is going to just take over everything we're all going to be out of a job in like about a year you know  um and MCP I think
(35:30) is similar right mcp is fantastic it's obviously as you said taking off as a standard which is probably a good thing you don't want to see a protracted war in standards necessarily um but yeah what what can it but but fundamentally it's it's still so nent in terms of what it can in terms of um it enables a lot of things but there needs to be a lot of infrastructure built around it there needs to be a lot of tooling built around it um you know how do you ensure how do you guard against tool confusion right when you give  an agent like hundreds of tools hundreds of APIs how
(35:59) does it know which one to pick which one to use at the right time how do you guard against hallucinations and so forth how do you make the MTPs more rich so that the uh the agent knows how to use it in the best way all those problems are things that need to be solved yeah security is also very important right like we've also we've already seen potential examples of of um poisoning within the workflow the pipeline um so security is going to be pretty big in this to to trust it explicitly end to end if you're giving an agent your credit card you want it to be you know
(36:28) trustworthy the whole way and security plays a big part of that so I feel like that's and to be if I can sort of circle it back to what Speak Easy is is is all about that's kind of where I see you guys coming into play as well if you can create something that's consistent efficient and secure you know that that's that's going to be a big plus for everyone looking to use the platform to get to where they want to be with MCP with API integration um and with agent to agent  absolutely yeah i mean um you've summarized it very well i think that the future here for speak easy isn't simply hey we've generated this
(37:00) MCP artifact there's going to need to be a lot of infrastructure built around the artifact around security and around solving the problems of like auditing and logging and um tool confusion and prompt templating and so forth and so how do you make the NTPs as useful as possible to your developers and so we are working on some exciting stuff around making agents and AI easier to embed as part of your API um infrastructure yeah  yeah and I think um just to finish we got a couple of minutes left i just want to point out that I think for people that out there that want to use the platform go ahead
(37:41) like I did sign up download it you know I'm on a Mac brew install bam it kind of comes up and and you're there already a little bit of authentication so I would just encourage people to have a play with it and is that are you targeting sort of that sort of developer community to start with or you know what what's your sort of target audience in terms of the use case for the platform at the moment in in a 30 seconds or so  yeah absolutely i mean our target audience are ultimately developers um our you know um Northstar is making it easier
(38:12) for APIs to be integrated with right to unlock the full potential of APIs um so you can just get started for free um hopefully in a very very short order with an open API spec um yeah  yeah and in terms of I think what I what I sort of experienced is that it's not going to be sort of straightforward as ways to to you're going to have issues with your API specs because that's the nature of them they're quite complex by nature so you were saying you've got a you've got an editor in there that helps you actually work through the problems and you can actually modify the spec to make
(38:41) it to make it not so complex and then it spits it out so everything that you're doing is about making things a lot easier along the way so I love that so in a minute or so just give us a bit of a future outlook for speak easy what's happen I made a joke earlier with you on the pre-show like you can't talk about years ahead let's talk about six months ahead because that six months is now the new five years  that's about as far as you can yeah predict um for the next six months speak easy is all about um making the tooling better so that the input
(39:11) open API specification the description of your API is better so it's the editor experience it's using AI there and it's around the agent tooling stack how do you make um AI easy to use within your internal software how do you use agents internally and externally as well awesome great stuff hey I think the value of speak easy has been articulated amazingly you're hitting all the points so I'm I'm looking forward to seeing where you guys go um again next 6 months 12 months 18 months you've got some great funding as well you've got some great backing so it's it's good lineage
(39:40) great work um thanks for being on episode 101 of Great Things with Great Tech  thank you so much Anthony been great talking to you  all right  hey just as a reminder thanks for listening to this episode stay tuned for more episodes where we continue to highlight companies and technology shaping our world don't forget to follow us on social media at gtwgtodcast and visit gtwgt.
(40:04) com for more great content and all past episodes if you enjoyed this episode make sure to subscribe on your favorite podcast platform and on YouTube please spread the word and if you feel like it drop a review thanks for joining us and we'll see you next time on Great Things with Great Tech [Music]