Permalink: 20251009101622

(…) when you put all your entire API surface and turn it into tools, the funniest thing that happens is first the LLM makes the wrong guess about which [MCP] tool to use. And once it realizes that it’s done the mistake, there’s like this sort of cascade of events where it tries to call so many other tools that it’s just not figuring its way towards a solution to the initial prompt or the initial problem.—Georges Haidar, August 2025

Permalink: 20251008141937

(…) it’s likely that the people who live in these houses don’t know their neighbors because all life either takes place inside the house or they get in the car and they go somewhere else. And there’s an exactly analogous picture to this in computation, which is this. These people don’t know their neighbors. They’re each in their own separate world. They’re not engaging with the people around them. And the people here and people here might be very social people, but the technology is structured in such a way that it’s inherently isolating, that it inherently pulls them apart.—Bret Victor, The Screenless City Conference, 2025

Permalink: 20251006104019

Found at “Why designers abandoned their dreams of changing the world” on 2025-10-06T10:40:19+02:00.

The product is less important than the process.

The most prestigious design schools now concentrate on narrative, on storytelling — the product (that word too would be frowned upon) is less important than the process. The result is a strange situation in which young designers are situated between activism, performative gesture and a residual urge to create which is sublimated into something between journalism, installation, anthropology, sociology and superstition. A large part of design is now a critical field, producing provocations rather than necessarily solutions. It is aimed somewhere else.

Designers become brands.

(…) the most famous designers do all they can to become brands in their own right. It is not their products which are the end result of their labours but themselves. (…) Big-name designers greedily accept commissions for huge, empty installations which aspire to the condition of art but end up more like experiential branding.

Walter Gropius house, © Getty Images. This house in Dessau, Germany, is an icon of Bauhaus design. It was on the original article to show an example of what good design looks like.

Permalink: 20251002105431

I’ve been thinking a lot lately about the notion of API prototyping and its relationship to API simulation. There are three areas worth spending more time exploring: a) “just-in-time” API prototyping, enabling experimentation during API Design; b) API simulation, enabling the creation of “high-fidelity” API mock-ups that feel like the real thing; and c) the programming language used to add simulation logic to an API prototype.

Permalink: 20250925094442

What is the meaning of trusting an API specification, e.g., OpenAPI? Is the trust related to the people who work on the specification? Is it related to the technical quality of the work? To both? Or something else?

Permalink: 20250924152809

Kin Lane just quoted me, sharing how I feel about the existing API directories.

Found at “Trust is the Secret Ingredient Missing in API Discovery by The API Evangelist” on 2025-09-24T15:28:09+02:00.

It’s because I can’t trust any of those places enough to accept that their results are the best match for what I need.

Permalink: 20250922181005

I’ve been researching the importance of workflow metrics. Here’s a short list of the most important metrics I’m considering:

• Execution success vs. failure rates (overall reliability).
• Average execution time (per workflow and per step/node).
• Throughput (workflows/minute or executions/day).
• Error frequency and error type distribution (identifying failing connectors, steps, or APIs).
• Resource utilization (CPU, memory, cost implications for high-volume workflows).
• Queue length / latency (how long items wait before processing).
• Retries and recovery events (resilience of workflows).
• Custom business metrics (e.g. number of invoices processed, leads captured, etc., tied to workflows).

This is quite an interesting area, definitely worth more exploration.

Permalink: On the complexification of API business alignment

I raise a red flag whenever I feel someone is making a simple topic feel complex. The last time this happened was a few days ago when an API industry veteran insisted on the complexification of API business alignment. Yes, the term “complexification” exists. It’s the process of increasing the complexity of something by splitting it into several parts. The goal is to make readers lose focus on the immediate problem and, instead, emphasize that the problem is in the lack of clarity of the outcome.

This is what happened when I was suggesting that the API industry could have an official registry and API search engine. I proposed that OpenAPI could launch it since it’s an official organization that consumers trust. The reaction was to show that such a registry would never work because it wouldn’t be able to align with the business needs, whatever they are. That is correct, an API registry would never be able to align with all existing business needs. In the same way, a Web page registry, a.k.a. a search engine, doesn’t align with all existing business needs.

There is, however, at least one business need with which an API registry aligns. But before getting to that, let me try to demystify what “business alignment” is. Well, in simple terms, aligning an API with the needs of a business means that the API supports the goals of the business, whatever they are. There’s nothing more to it. If the goal of the business is to reduce costs, then the APIs should help automate manual processes. If the goal of the business is to attract partners, then the APIs should help those partners easily exchange data with the company. What happens if the goal of the business is to attract new customers?

If the goal of the business is to attract new customers, then the APIs should be easy to discover and onboard. Let me repeat and emphasize. If the goal of the business is to attract new customers, then the APIs should be easy to discover and onboard. How can we make APIs easy to discover by potential customers? One way is to publish information about those APIs in a place where potential customers can go to and find them. This place is the OpenAPI registry I was proposing.

There is no complexity in this. It’s a simple alignment between one of many possible business needs with something that generates the desired goal. What businesses do with those new customers, how they onboard them, and, among other things, the pricing strategies they have in place, are beyond what the OpenAPI registry would offer.

Permalink: 20250918093507

I’ve been chatting with David Biesack, the Chief API Officer at Apiture, a digital banking company in business in the US since 2017. Apiture offers a set of digital financial features to banks through APIs. They also connect financial partners to banks through their platform. Biesack shared that they onboard new partners to use their APIs manually. They do have a public developer portal that potential partners find through Google searches. They use the information on the developer portal to educate partners and their developer teams on how to integrate with their APIs.

Permalink: 20250917180123

I just had a chat with Bruno Amaral, the founder of Gregory AI, a medical research assistant software. Amaral runs a version of the assistant dedicated to multiple sclerosis. This assistant offers information about research articles and their authors. Amaral told me that he wishes there were an API directory so he could publish the assistant API there. For now, he has to contact developers directly to show them how the API works. There are two other research projects based on the same technology that could benefit from publicizing their APIs properly. One is dedicated to brain regeneration, and the other one to encephalitis (in Portuguese, mostly).

Permalink: 20250917100756

I recently had a chat with Vinicius Marques, a Brazilian banking technology professional, about what it means to be an API producer in his field of expertise. Initially, I just wanted to know what the banking industry is doing to attract consumers to their APIs. What he told me was quite interesting. In general, banks have a public API portal where developers can see what APIs are available. The public portal also has API documentation, which, according to Marques, is quite hard to keep up to date. Most developers find those APIs through Google searches. Some of them get in after consulting the Brazilian Central Bank API directory.

Permalink: 20250916142116

Here’s a list of API directories I’ve been profiling:

• APIs.guru
• Public APIs
• API list
• Postman API Network
• APILayer Marketplace
• APIsList
• APIs.io
• API Expert
• API Tracker

Permalink: 20250916123408

Found at “How to Promote and Market your API: API Directories | Zuplo Learning Center” on 2025-09-16T12:34:08+02:00.

API directories (also known as API indexes) are essentially lists of APIs that developers will find useful when solving certain problems. Imagine you needed an API for sending emails. Sure, you could just type that into Google - and you’ll find options like SendGrid and Mailgun - but you might miss out on other APIs like Resend that are potentially more affordable or easier to use. These days, most established APIs have the resources to invest heavily in ads and SEO, while many startups do not - essentially leaving you with fewer options. API directories act as (sometimes) neutral platforms where all APIs get indexed, are discoverable, and are presented similarly.

Permalink: A registry for APIs?

I recently wrote a post on LinkedIn sharing that OpenAPI could follow what MCP is doing to solve discovery. MCP introduced the registry on September 8, 2025. Their goal is to centrally “provide MCP clients with a list of MCP servers, like an app store for MCP servers.” It sounds like a fantastic approach and something I think we can replicate for APIs.

The reactions to my LinkedIn post were not at all encouraging. Reading those comments makes you feel like many people don’t want APIs to be easily discoverable. It’s interesting to see the pushback by some API industry veterans and also the arguments being used.

One comment mentions that “the business alignment is the problem.” It also mentions the failures of some API directories like ProgrammableWeb, Mashape, RapidAPI, and Postman Network. First of all, there’s a distinction to be made between API directories (or registries) and API marketplaces. ProgrammableWeb and Postman Network can be considered API directories. On the other hand, Mashape and RapidAPI are API marketplaces. The difference is that marketplaces manage the interaction of consumers with the listed APIs.

So, what happened to ProgrammableWeb to fail? It started in 2005 as an API-focused blog and a directory of APIs. ProgrammableWeb changed ownership a few times. It was sold to Alcatel-Lucent in 2010. ProgrammableWeb founder moved to Alcatel-Lucent and kept running the blog and API directory as an independent organization. Alcatel-Lucent mentioned that the reason behind the acquisition was to help foster an ecosystem of service providers. At the time Laura Merling, VP at Alcatel-Lucent said that “If you look at any organization that launches an API, you quickly realize that the one thing the most successful APIs have in common is a vibrant developer ecosystem.” Then, in 2013, ProgrammableWeb was sold to MuleSoft. MuleSoft had recently launched APIhub, an API publishing platform, and wanted to join forces with the growing API directory that ProgrammableWeb provided. Then, in 2018, MuleSoft was sold to Salesforce for $6.5 billion. ProgrammableWeb kept running until 2022, when it stopped making sense to keep a large manually curated API directory. Did ProgrammableWeb fail? I don’t think so. It lasted 17 years, from 2005 to 2022. It kept growing and growing, reaching 1,000 APIs in July 2008, and more than 25,000 APIs by the time it shut down. It was clear that the system had to change from being manually curated to something else.

What about Postman Network? Did it also fail? Let’s find out. It was launched in 2017 as a curated public directory of APIs and Postman collections. I started with a few hundred APIs, curated by Postman and its partners. In 2018, Postman introduced its private API network as a way to give companies a private way to publish and discover their APIs. This feature was well received and helped grow the public API network as companies themselves began contributing. In 2021, Postman added more community-focused features to the public API network, making it grow in the number of APIs. In 2022, it had thousands of APIs, in 2023 it reached 75,000 APIs, and in 2025 it reached more than 100,000 APIs. It looks like the growth lever was opening it to contributions from the API owners themselves instead of keeping it closely curated. Postman Network is active and growing at the time of this writing. It doesn’t look like a failure to me.

What about Mashape? Mashape wasn’t an API directory, so I’m not going to do a full analysis. Mashape started in 2009 with the goal of becoming an API marketplace. At its peak, in 2015, it listed over 15,000 APIs. It was also in that year that Mashape open-sourced the API gateway it used to manage the billions of monthly API requests, serving over 150,000 developers worldwide. Kong, its API gateway, became its main focus in 2017. Mashape sold its API marketplace business to RapidAPI and rebranded to Kong, focusing on the API gateway and management solutions. Was it a failure? I don’t think so.

And what about RapidAPI? It’s also a marketplace, not a directory. After it acquired Mashape’s API marketplace in 2017, it became the largest inventory of ready-to-use APIs with over 7,500 APIs. It served over 370,000 developers, making over 300 billion requests per month. At its peak, in 2023, RapidAPI, then rebranded to Rapid, hosted over 40,000 APIs. After having company management issues, it was sold to Nokia in 2024. It could have grown much more, I’d say. The company management lost focus and spent too many resources going after markets that didn’t scale. In that sense, we can say that it failed.

After doing this analysis, it looks like the approach that scales is the one followed by Postman. Letting API producers themselves publish their APIs is key.

Permalink: 20250911133843

Found at “RSL Open License Protocol (OLP) 1.0 Draft Specification | RSL” on 2025-09-11T13:38:43+02:00.

OLP introduces a new OAuth 2.0 grant type, rsl, to support using RSL licenses as credentials for controlling access to digital assets.

Permalink: Sponsorship vs. Advertising

Most people I know confuse sponsorship with advertising, so I’m putting together this short guide to help me remember what the differences are. I didn’t invent anything. Instead, I’m getting inspiration from people who have a lot of experience in the field of marketing. Let me start by exploring the different types of advertising, according to their goals.

• Awareness (broad reach, brand recognition): Make people know the brand or product exists. Works well on ads that focus on the product image.
• Consideration (product education, credibility): Get people to think about the brand, compare, and build interest. Works with sponsored blog posts explaining what the product does and how it works.
• Conversion (action-driven, measurable ROI): Drive immediate action, such as signing up or making a purchase. Works well with Google search results ads and email campaigns.
• Retention (customer relationship nurturing): Keep existing customers engaged and buying again. Works well with drip marketing and mobile push notifications.
• Advocacy (word-of-mouth, community leverage): Turn customers into promoters. Works well with user-generated content and influencer campaigns.
• Recruitment (employer image building): Attract talent, not customers. Works well with LinkedIn recruitment ads and employer branding campaigns.

Let’s now look at the different types of sponsorship, according to their goals.

• Brand Awareness (high reach, logo placement, name recognition): Reach large audiences, make the brand visible. Works well with stadium naming rights, music festival headline, and media “brought to you by.”
• Reputation (association with credibility, quality): Improve perception, build prestige, align with values. Works well with art exhibitions, tree planting, clean energy, and sports.
• Consideration (hands-on experiences, product showcases, and content integration): Educate and build preference so customers consider the brand. Works well with hackathons, podcasts, blogs, webinars, and sampling at trade shows.
• Conversion (performance-linked, call-to-action oriented): Drive purchases, sign-ups, or immediate revenue. Works well with influencers, co-branding, and free product samples.
• Loyalty (exclusive access, VIP experiences, and rewards): Strengthen relationships with existing customers and reduce churn. Works well with exclusive brand lounges for members at airports and sports arenas, fan clubs, and partnerships with community events.
• Advocacy (grassroots, values-driven, and user participation): Turn customers and communities into brand advocates. Works well with local sports teams and schools, user-generated content contests, and social good projects.

As you can see, there are overlapping goals between advertising and sponsorship. However, they’re different in the way you execute the campaigns. Let’s now see a comparison matrix where I measure the differences across eight aspects.

Permalink: 20250910101448

Found at “visual frameworks – A language of patterns” on 2025-09-10T10:14:48+02:00.

Visual frameworks are patterns to help you think creatively, reframe challenging situations, and imagine possible strategies and solutions.

Infinite Loop Visual Framework. Are you stuck in an infinite loop, oscillating between two extremes, when what you really want is the happy medium? Can you improve your understanding of the oscillating pattern by diagramming it out? Can you design a more measured, managed approach that helps you escape the infinite loop?

Permalink: 20250909171051

Here’s a list of evals tools and frameworks I’ve been profiling:

• OpenAI Evals: Open-source, from OpenAI.
• LangSmith: Observability and Evals.
• PromptPex: More focused on prompt testing.
• ChainForge: Prompt robustness testing visual UI.
• PromptLayer: Full end-to-end AI testing and monitoring.
• Garak: Security-focused, from NVIDIA.

ChainForge example flow. In this example ChainForge evaluates model robustness to prompt injection attacks.

Permalink: 20250909102344

Found at “Introducing the MCP Registry” on 2025-09-09T10:23:44+02:00.

Discovery is probably one of the hardest things to get right. It looks like MCP is doing things the right way by showing others how to promote the discovery of MCP servers.

By standardizing how servers are distributed and discovered, we’re expanding their reach while making it easier for clients to get connected.

Permalink: 20250905104453

Found at “Comparative advantage - Wikipedia” on 2025-09-05T10:44:53+02:00.

Comparative advantage in an economic model is the advantage over others in producing a particular good.

Permalink: 20250904153216

Last month was very quiet here, not because I wasn’t doing anything, but because I chose not to publish here as much. Here’s a summary of different things I published in August 2025 elsewhere:

• Full API MCP Exposure vs. Specific Workflows on August 29, about the differences between exposing a full API through an MCP server in comparison to picking a handful of workflows.
• Machine-Oriented Use-Case Documentation on August 22, exploring ways of making multi-step API use cases understandable by AI agents.
• Enterprise MCP Authorization on August 15, about the choices you have to add authorization to an MCP server.
• Connecting AI to APIs in a Standard Way on August 13, showing how you can use API unification in combination with MCP to make it easy to connect AI to external tools.
• Documenting Your API Around Its Capabilities on August 8, about the importance of documenting capabilities in making consumers understand what an API does.

Permalink: 20250731142030

This is how I write, in five simple steps:

1. The problem: show readers that there’s a serious problem that needs to be fixed.
2. A possible solution: explore one or more possible solutions to the problem.
3. Solving the problem: tell the story of how to solve the problem using one of the possible solutions.
4. Drawback: show that the solution is not perfect, after all, and present its disadvantages.
5. Final decision: decide if readers should stay with the problem or opt for the proposed solution.

Permalink: 20250730094806

Found at “Malleable software in the age of LLMs” on 2025-07-30T09:48:06+02:00.

Why might LLMs be a big deal for empowering users with computation?

Permalink: 20250728174123

My practice is led so much by not fully understanding or not fully knowing what the next step is going to be.—Lotus Kang

This definitely reminds me of this quote by Philip Guston: /2024/06/10/20240610152531.

Permalink: 20250724223100

Found at “GitHub Spark - Dream it. See it. Ship it.” on 2025-07-24T22:31:00+02:00.

GitHub Spark helps you transform your ideas into full-stack intelligent apps and publish with a single click.

Permalink: 20250724173536

If the definition of a polyglot is “a person who knows and is able to use several languages,” then I’m definitely one. Not just that, but I have to use several languages on a daily basis just to go through the day. At home, I use Portuguese, at work, I use Spanish and English (both written and spoken), and in the street, I use Spanish. Even though I have lived in Spain since August 2011, I never took the time to learn proper Spanish simply because I never needed to. Until last year. Where I now work, most people speak only Spanish, so communicating without using it would be a major barrier. Every single day, I learn a few words in Spanish and some more work-related expressions. What usually happens is that I get a heavy load of what I call “cognitive deficit,” which is the lack of ability to think properly after a few hours of using a language that still doesn’t feel native.

Permalink: 20250722143439

I’ve been thinking a lot about API prototyping lately. I feel that prototyping has to assume a key role in the API lifecycle. The participation of stakeholders in the API design process is something that benefits the quality of the final result.

Permalink: 20250721095658

Found at “Introducing XMLUI” on 2025-07-21T09:56:58+02:00.

(…) XMLUI brings the VB model to the modern web and its React-based component ecosystem. XMLUI wraps React and CSS and provides a suite of components that you compose with XML markup.

Permalink: 20250718142336

Readymades, or objets trouvés in French, are items artists find that aren’t normally used to produce art. Artists use those items to create art pieces such as collages in visual art and sample-based music.

Using readymades was popularized by artist Duchamp. Three examples of popular Duchamp readymades are the “Fontaine,” the “Porte bouteille,” and the “Roue the bicyclette.” The “Fontaine” is, in fact, a urinal. However, the artist calls it a “fountain.”

Permalink: 20250718135303

Found at “From Beta to Bedrock: Build Products that Stick.” on 2025-07-18T13:53:03+02:00.

Bedrock is the core element of your product that truly matters to users. It’s the fundamental building block that provides value and stays relevant over time.

Permalink: 20250718093949

Found at “All AI Models Might Be The Same” on 2025-07-18T09:39:49+02:00.

(…) there is only one way in which things are related, and this comes from the underlying world we live in. Put another way, our brains build up complicated models of the world in which we live, and the model of the world that my brain relies on is very similar to the one in yours.

Permalink: 20250717144151

Found at “From Memo to Movement: Shopify’s Cultural Adoption of AI | First Round” on 2025-07-17T14:41:51+02:00.

(…) the entire senior leadership team needs to agree that AI adoption is the most important thing you can do, which includes legal teams. Alignment at the highest level means everyone understands you have to find a way to get to “yes,” including the key conversations around security and privacy.

Permalink: 20250716153337

Found at “Infinite Mac” on 2025-07-16 15:33:37 +02:00.

Infinite Mac is a collection of classic Macintosh and NeXT system releases and software, all easily accessible from the comfort of a web browser.

Mac System 1.0. Initial system software release, shipped with the Mac 128K on January 24, 1984.

Permalink: 20250715125740

Found at “Focusing on Capabilities Is a Win - by Mike Amundsen” on 2025-07-15 12:57:40 +02:00.

Describing your API as a set of available operations, each with names, purposes, expected results, and optional constraints, gives intelligent clients the raw materials they need to plan and adapt.

Permalink: Sunshade

Sunshade. A contemplative illustration of the feeling you get when the sun rays hit your skin.

Permalink: 20250707124857

Try not to get a job! That doesn’t mean “Try not to do anything.” It means “Try to leave yourself in a position where you do the things that you want to do with your time."—Brian Eno

Permalink: 20250611094235

Found at “The Gentle Singularity - Sam Altman” on 2025-06-11 09:42:35 +02:00.

The singularity isn’t a new concept. In 1999, Ray Kurzweil famously created the law of accelerating returns. Its gist is that the rate of change in evolution gets to a point where it increases exponentially. It’s easy to believe that, in the exponential growth period, everything will be better, faster than before.

(…) the gains to quality of life from AI driving faster scientific progress and increased productivity will be enormous; the future can be vastly better than the present.

Then it gets to a point where it’s practically impossible to control and gains a life of its own—the singularity.

It’s hard to even imagine today what we will have discovered by 2035; maybe we will go from solving high-energy physics one year to beginning space colonization the next year; or from a major materials science breakthrough one year to true high-bandwidth brain-computer interfaces the next year.

Ideas become more important than execution. Until now we’ve always been saying the opposite, that ideas really don’t matter that much. What really matters is how you execute those ideas.

For a long time, technical people in the startup industry have made fun of “the idea guys”; people who had an idea and were looking for a team to build it. It now looks to me like they are about to have their day in the sun.

Permalink: 20250605091402

Found at “Introducing the HubSpot deep research connector” on 2025-06-05 09:14:02 +02:00.

That’s it—no API configurations, no dev work. Once connected, you can start using natural language to analyze your customer data with deep research, and take action on those insights to boost marketing, sales, and service outcomes.

Permalink: 20250603151843

Found at “Good Writing” on 2025-06-03 15:18:43 +02:00.

An essay is a cleaned up train of thought, in the same way dialogue is cleaned up conversation, and a train of thought has a natural rhythm

Permalink: 20250521155438

Found at “Power Hungry: AI and our energy future | MIT Technology Review” on 2025-05-21 15:54:38 +02:00.

It’s well documented that AI is a power-hungry technology. But there has been far less reporting on the extent of that hunger, how much its appetite is set to grow in the coming years, where that power will come from, and who will pay for it.

Permalink: 20250518161739

Found at “The Impact Matrix: Moving to the golden quadrant | Seth’s Blog” on 2025-05-18 16:17:39 +02:00.

Tactics are tempting. We can lean into them, invest, build our skills and count on results.

Strategies are more elusive. And a mismatch between strategy and tactics leads to wasted effort.

In this 2 x 2 grid, you can see how easy it is to get stuck.

The impact matrix, by Seth Godin. This matrix shows you how picking the right strategy and coupling it with the right tactics can produce the impact you’re looking for.

Permalink: 20250515093455

Found at “Internet Artifacts - The Hacker’s Dictionary” on 2025-05-15 09:34:55 +02:00.

AUTOMAGICALLY adv. Automatically, but in a way which, for some reason (typically because it is too complicated, or too ugly, or perhaps even too trivial), I don’t feel like explaining to you. See MAGIC. Example: Some programs which produce XGP output files spool them automagically.

Permalink: 20250509181610

Found at “ALICE detects the conversion of lead into gold at the LHC | CERN” on 2025-05-09 18:16:10 +02:00.

Transforming the base metal lead into the precious metal gold was a dream of medieval alchemists.

Permalink: 20250508155318

Found at “Why Does Switzerland Have So Many Bunkers? — The Dial” on 2025-05-08 15:53:18 +02:00.

the European Union issued official statements urging residents to keep an emergency stockpile containing 72 hours’ worth of supplies on hand at all times. The exposure to war and man-made disaster feels more acute than it has since any other time since the end of the Cold War.

Permalink: 20250507141707

Found at “New smartphone labels for battery life and repairability are coming to the EU | The Verge” on 2025-05-07 14:17:07 +02:00.

The European Union has announced details of new mandatory labels for smartphones and tablets sold in the bloc, which include ratings for energy efficiency, durability, and repairability. Hardware will also have to meet new “ecodesign requirements” to be sold in the EU, including a requirement to make spare parts available for repair.

Permalink: 20250507140910

Found at “mobygratis - Free Moby music to empower your creative projects” on 2025-05-07 14:09:10 +02:00.

mobygratis exists for one reason; to provide free instrumental music for creators. any creators. all creators; filmmakers, musicians, students, influencers, choreographers, non profits, video editors, remixers, singers, gamers, animators, rappers, etc etc.