Alejandro Legarda

I build AI systems that follow rules, adapt, self-correct, show their work, converge, never guess and self-improve.

What I do

  1. Hybrid: AI and code work together to create powerful and reliable systems.
  2. Auditable: Designed to be traceable and understandable.
  3. Convergent: Statistical techniques pull reliable answers out of unreliable models, lifting output quality past what any single model can reach.
  4. Self-healing: When an output fails a check, the system pinpoints the fault, repairs and confirms the fix before moving on.
  5. Calibrated: The system knows how sure it is. Confidence decides how much scrutiny a case gets, and anything still uncertain goes to a person.
  6. Self-improving: Every run becomes data for the next version. Every failure is learned from and remembered.
The loom selects coloured threads from a pattern card and weaves stars, spirals, braids and geometric mosaics. Drafts gain ink as independent checks pass. Each pass meets a new fault: a knot in a source yarn that drags the cloth into a V, a jammed selector that leaves a diamond-shaped hole, or the two models' yarns splitting apart. A carriage pulls the knotted yarn, darns the hole or presses the split together, with a person settling the disagreement, and the pattern is restored. A return thread records the fault's shape on a woven memory tape. In the next pass the same fault comes back in the same place. The loom reads the record and fits hardware that stops it before it is woven: a clip holds back the knot, pins lift the jammed selectors, or a guide thread keeps the yarns together. The pattern weaves through whole, and the fitted hardware stays on the loom. This is an illustration of the approach, not live system activity.

Things I’ve done

ClauseFlow

AI engineering lead at DAC, since 2024

I lead AI development at DAC and built the AI engine behind ClauseFlow. It compiles a credit agreement into an auditable data layer that makes frontier models better at legal comprehension, so they follow how provisions work together across hundreds of pages.

Architecture

Designed an 81-step pipeline where adversarial agents cross-examine every finding against the agreement text, repairing or withdrawing it before anything ships.

What-if

Built an engine that runs what-if scenarios on credit agreements, tracing how a single event cascades through the deal and outputting an auditable clause sequence.

Discovery

Finds liability management maneuvers no one has published, the kind that have cost lenders billions, by chaining clauses that each pass review on their own.

Demos

01Vox Transfer

Released · Speech to text for macOS and Linux

Your voice.
In any app.

Tap a key, speak, tap again. Your words appear wherever you’re typing.

macOS & Linux · OpenAI or fully offline

Get Vox Transfer Open source
Speak naturally. Write anywhere.

In this Notes example, say “The best ideas don’t wait for a keyboard. Capture them before they disappear.”. Vox Transfer inserts the sentence at your cursor.

02Loremaster

Feeds the AI Game Master

Asked with no libraryPaint Saint Ilsa of the Lanterns.

Saint Ilsa is traditionally shown in flowing white robes1 beneath a halo of golden stars2, her lantern raised high3 over a stormy sea4, with a crown of river reeds5 in her hair.

500k+passages

500+books

Served to any LLM over MCP

Point at a bookTap a book

search_lore

how the lantern saint is shown

Try
She is never shown with her lantern raised. It hangs low, at the height of the water, for she was searching for the living and had no business with the sky. She went barefoot on the roofs, and the sisters of her house crop their hair short in her memory to this day.

Source: Lives of the River Saints, Heading: Ilse of the Lanterns

read_source read on from the hit

For three nights the river stood in the lower town, and for three nights a single light was seen moving across the roofs.

Source: The Flood Registers of the Lower Town, Heading: The Year of High Water

The Boatmen paid for the east window of her chapel, and would have no colour in it but the blue of deep water and the yellow of a lamp.

Source: The Guilds of the River Towns, Heading: The Boatmen

Asked with the libraryPaint Saint Ilsa of the Lanterns.

Complete knowledge, every word sourced

I wanted an AI that knows everything about one subject and can prove every word it says. An LLM can’t do that on its own. It answers from a fallible memory of its training, and at best it checks itself against whatever a web search turns up. It sounds just as sure when it is wrong as when it is right.

Private repository.

Confident, and wrong

Asked to paint a saint from a fantasy world, an LLM answers at once. Five of its details appear in no book, and nothing in the answer says which.

So I gave it the books

Every passage keeps its book and heading, so every quote can be traced to its source. Each dot here is one passage.

No claim without a quote

It searches by wording and by meaning, then opens the book and reads on. Every claim it makes must carry a quote.

Asked again

This time my illustration skill looks her up first, then designs every pane from what the books say. Pick a pane or its label to read the passage behind it.

The saint, her books and every passage on this page are stand-ins written for it. The real library is private.

03AI Game Master

Builds its world and stories from the Loremaster

Night falls over the river docks. The last barge is tying up, and the tavern on the quay still has its lamps lit.

What do you do?

  • IntrigueI follow the tax collector home to see who he pays.
  • BattleI hold the bridge while the cart gets across.
  • FaithI pray at the shrine until my sister’s fever breaks.
  • CareerI quit the army to apprentice with an apothecary.
  • MagicI cast the spell again, even though it went wrong.
  • MoneyI bank my savings and pray the house doesn’t fail.
  • WoundsI let the surgeon take the arm before the rot spreads.
  • TravelI take the next barge downriver and start over.
  • CrimeI pick the counting-house lock once the watch passes.
  • SecretsI find out what the priest is hiding.
  • FearI stand my ground as the thing comes out of the fog.
  • AmbitionI marry into the family that ruined mine.
  • StudyI spend a season reading in the temple library.
  • RevengeI hunt down the man who hanged my brother.
  • TemptationI listen to the voice that keeps offering to help.
  • HuntingI track the thing that took the miller’s boy.

All of it in one game

  1. You come in out of the rain. The tavern on the quay is warm and half full, and nobody looks up. At the counter a woman in a mended coat slides a ring toward the innkeeper. He turns it over once, shakes his head and slides it back.

  2. Stranger

    Please. It’s real gold, and I only need two crowns for the barge at dawn.

  3. I ask if I can see the ring, and hold it up to the lamp.

  4. Skill rollEvaluate

    Need60+

    Judging what the ring is really worth.

    Rolled 72. Needed 60, so you see what it is.

  5. Stranger

    It was my mother’s. I wouldn’t sell it if I had any other way.

  6. It is worth fifty crowns at least. Inside the band, worn almost smooth, is the seal of the counting house on the square.

I give her back the ring and two crowns, and ask how she came by it.What do you do? Send

Session 14

  1. The harbourmaster opens the door and waves you in. Forgets

    This game The harbourmaster drowned in session 2. His widow answers the door.

  2. Inside, a chest holds ten thousand gold crowns. Invents

    This game Inside are a purse and a ledger, the same as when you first looked.

  3. You slip past his guards without a sound. Skips a roll

    This game Two guards stand at the door. Roll Stealth to slip past them.

  4. Back at the inn, the keeper calls out, “Welcome back, Brandt.” Mixes up who knows what

    This game “Welcome back, Kessler.” She only knows the name you gave her.

The rulebook, in code
  • MagicSpells that can go wrong
  • CombatOn a tactical grid
  • WoundsInjury, infection, disease
  • FaithPrayers and divine wrath
  • CareersSkills gained by advancing
  • DiceRolled in code, never by the LLM
Meanwhile, in story time 06:00
  1. Hills

    The warband in the hills breaks camp and turns toward the river road.

  2. Bank

    The bank holding your savings closes. Its partners left by boat at dawn.

  3. Temple

    The priest you questioned burns a ledger and sends a boy off with a letter.

  4. Gate

    A rider brings word of the warband. The captain shuts the east gate.

  5. Inn

    A bailiff asks the innkeeper who the woman with the ring was talking to last night.

Nobody in town waits for the player

Creating a new kind of game

A human game master can answer anything a player tries, which is why tabletop roleplaying never fully made the move to videogames, where every choice had to be scripted in advance. This game puts an LLM in that chair. The campaign is played in plain language and held to the printed rules by a deterministic engine, and between scenes its people keep pursuing plans of their own, so the world feels real and the story unfolds around whatever the player creates.

Private repository.

Stories only a table could tell

In a tabletop roleplaying game, a player can live any life its world allows and try whatever plan they can put into words, and a game master turns it into play. Game makers have chased that freedom since the first home computers, and the games that came closest are among the most celebrated ever made. Each one still ends where its writers stopped.

A game made of language

At a real table a player can say anything at all, and the game master decides what it means for the world and when it calls for a roll. Play happens in language, so the possibilities never run out, and no program written ahead of time can keep up with them. For the first time, an LLM can take that chair.

Long campaigns drift

Over dozens of sessions an LLM on its own loses track of what has happened. Dead men come back, and rooms fill with treasure that was never there. This game holds its world together for the whole campaign.

Played by the book

Everything the rulebook governs plays by its rules. A spell can go wrong and leave a taint that builds over a campaign, and a wound left to fester can turn into disease weeks later. Every roll is made in code, and an automated check holds each rule to the page of the book it cites.

A world that goes on without the player

Offstage, the whole town keeps going. Every person in it has money and plans of their own, and news travels only as fast as someone can carry it. The design borrows from simulation games as much as from the tabletop, and what comes out is a kind of game that has never existed before.

04TRACE

A personal news agent · In development

I wanted news I actually care about.

So I built a news aggregator my personal AI agent can tailor to me. It starts from what I’m working on and what I care about.

Your agent already has the context.

It draws on what it knows about you, your projects and notes, and available browsing history to build a reading profile. TRACE judges incoming stories against that profile. Your agent can revise it as your interests change.

Click to pin · Double-click to expandTap to pin · Expand to read

Microsoft Research
Diagram of Project Quine’s closed-loop approach to biological research. A scientist asks questions that are passed to Quine for in silico computation. Quine combines a world model of biology, spanning genomics, proteins, chemistry, cell state, and bioimaging, with a harness for orchestration and reasoning, supported by knowledge and tools. Quine generates proposals that are tested through real-world wet-lab experiments. Experimental measurements return to the scientist and inform the next question, creating an iterative loop between computational modeling and real-world experimentation.

Introducing Quine: An AI research system designed for the complexity of biology

At a glance

  • Quine (opens in new tab) is a research effort to create a multimodal world model of biology and an interactive harness connecting models, scientific tools, literature, and researchers. 
  • In collaboration with researchers at the Broad Institute of Harvard and MIT, we have used this system to prioritize compounds predicted to drive therapeutic tumor-state shifts and validated several top-ranked candidates across multiple wet-lab assays. 
  • The Quine Fellows program (opens in new tab) will give a cohort of scientists access to the system and an opportunity to accelerate their own research and provide scientific feedback.  
  • Quine is experimental research technology intended only for research, not clinical or medical use, and its outputs may be incomplete or inaccurate and require review by qualified researchers and appropriate scientific and experimental validation. As the technology matures, we expect to expand access through products like Microsoft Discovery (opens in new tab).

For more than two decades, Microsoft Research has worked at the intersection of computation and biology. Our research has spanned immunology, virology, genomics, biomedical imaging, cell biology, and protein engineering. That work has produced foundational methods, new science, and technology that reached the clinic, from rare and infectious disease diagnosis to cancer biomarker detection.

Across that work, one lesson has become increasingly clear: biology does not divide itself into the neat boundaries our models and tools often do. Genes influence proteins; proteins interact within cells; cells organize into tissues; and experiments continually reshape what scientists know and what they choose to ask next. Making progress on the hardest biological questions therefore requires more than increasingly capable models of individual datasets or tasks. It requires systems that can connect knowledge across scale and modalities, reason about experiments and evidence, and participate in the iterative process through which science advances. 

Today, Microsoft Research is introducing Quine (opens in new tab), a research effort designed to work across those boundaries, reflecting our long-term vision for a discovery system that evolves through scientific use. Quine brings together a world model of biology with a harness that connects scientific tools, literature, the wet lab, and the researchers using them.

The limits of experimentation

Even as experimental techniques have improved and wet-lab throughput has increased, biology remains fundamentally constrained by time and complexity. Nature cannot be rushed, nor can it be derived from first principles. Experiments are slow, iteration cycles are long, and many of the most important questions involve interactions, combinatorial design spaces, and downstream effects that are simply too large to explore experimentally alone.

At the same time, advances in large-scale machine learning, particularly the emergence of general-purpose foundation models and reasoning models that can iteratively work through problems, suggest a new possibility. These systems are beginning to demonstrate capabilities beyond pattern recognition: integrating information across domains, reasoning over abstractions, and supporting iterative problem-solving. Just as importantly, many of the techniques developed for human language have proven remarkably adaptable to aspects of biology, enabling models to learn representations of biological systems across diverse data types and scales.

Continue reading at Microsoft Research ↗

LessWrong

Expansion in the Fermi paradox isn't a freebie

The Fermi paradox is premised on expansionism being intrinsic to civilizational development, most explicitly in Hanson's grabby aliens model. There's a Great Filter provided by expansionism itself that doesn't center on scientific and technological progress that I think is worth exploring. It's about the destabilizing nature of expansion events and the precarity of an expansionist disposition, and survives the standard counter that civilizational natural selection maximizes expansionism in the limit. It's a cosmic-scale cousin of habryka's Do not conquer what you cannot defend, where the concern is defending what's taken. Mine is about integrating it.

Questioning expansionism

Before and unrelated to that, a brief skepticism about the premise feels due and will be useful context. I don’t really believe that expansionism is a necessary civilizational posture, in particular in the long-arc limit relevant to Fermi time scales, and might be a rare/never phenomenon.

The argument for it seems to be from intuition, positing that selection pressure made organisms expansionist in the sense of always consuming what resource is available to support population growth. That tendency inheres genetically, hence something something cultures and civilizations must be like that on balance/in great number and also in the limit.

Recent memory certainly aligns. Technological progress has been regularly opening new avenues for resource acquisition and human civilization acts on them.

I don’t buy it. Cultures throughout history have not been uniformly expansionist and the rapid and steady influx of low hanging tech progress acts as confounder to our intuitions, which are based on recent times. We’ve seen that pattern for a long time, therefore it’s always like that.

But, counter to that, an awareness of the externalities of expansionist tendencies is already normative — the collective has metabolized that wielding technology to absorb newly accessible resource can have substrate-undermining externalities. This is a recent trend too and my reasoning here comes from intuition also (hence has that same recency confound), but I think this latter intuition has better foundations.

Externality awareness is not a naked observation about sociocultural trends. It fits into developmental models, which more or less all posit that a growing sphere of concern, a receptivity and saliencing of context and externalities, is constitutive of development. This observation is thus structural in kind, matching models of the development of other intelligent systems, whereas the linear intuition that “replicators expand therefore all the higher order structures constituted of them will tend to do the same” ignores that structural regularity. Also and to a lesser degree, that linear intuition is undermined/countered by noting that civilization comes to persist exactly by civilizing older base energies and drives. Take what you can for yourself, the atom of expansionism, is about the most primitive force at the individual level that coordination supportive of civilization suppresses. So I don’t think these two intuitions are on equal footing. A growing enacted concern for externalities has a precedented and principled significance.

Assuming expansionism

Continue reading at LessWrong ↗

Cloudflare Blog

We tested our own WAF with frontier AI models. Here’s what we found

“Is your WAF ready for frontier AI models?” We keep hearing this question from our customers, so we decided to find out.

When it comes to exploiting applications, what LLMs are really good at is iterating and mutating attack payloads faster than any human hacker could do. LLMs can use real-time responses to iterate and change their techniques by, for example, testing different encodings, sending the payload in a different part of the HTTP request, or moving to the next vulnerability to test.

Even before LLMs were around, security engineers used two common approaches to test applications: static and dynamic application security testing. The former analyzes code without executing it to identify vulnerabilities, while the latter probes running applications to find runtime flaws. There are plenty of works scanning code with frontier AI models, including details on how to build your own harness.

For the project described in this blog post, we took a dynamic approach: making the LLM act as if it was a hacker to evaluate whether a WAF is doing its job. The LLM had no visibility into source code, no view of the WAF's rules, and could only see selected HTTP response data.

We built a WAF tester that starts from known exploits and then iterates by changing how it is encoded or delivered, sends it again, and uses the response to choose the next variation. A request that was not blocked became a lead for human review, not a confirmed exploit.

We ran the tester against an authorized customer staging environment across six attack categories and recorded 1,107 attempts. After reviewing the non-blocked requests and removing malformed, benign, duplicate, and out-of-scope observations, the vast majority of the attacks were blocked by the Cloudflare WAF. The requests that got through helped us create new detections to harden our security to benefit all Cloudflare customers.

Here we will explain how we set up the system, the types of attacks we tested, which attack vectors bypassed the WAF more easily, and how we fixed it. Most importantly, we share what we learned from this process and how this exercise is becoming a foundational building block of our WAF development lifecycle.

Finally, we offer guidance to help you correctly deploy your WAF in front of your application and, most importantly, patch your software. A payload that bypasses the WAF still needs an exploitable application to succeed, so keeping your stack up-to-date remains one of the strongest defenses against attackers.

How the adaptive loop works

To test our WAF with frontier models, we built a system that iterates over multiple scenarios. A scenario means choosing one attack category, placing the input in a specific part of the request, starting with a version the WAF already blocked, and giving the tester a fixed number of attempts to try other variations. The loop runs LLM models twice: the first is the proposal call, the second is the review call.

The first call receives the starting request, the context, a short history of earlier results, and suggests the next variation, then the code builds and sends the request. The review call receives the request context, response status, selected headers, and the response body. The loop stops when mutations stop producing useful variations or when a hard coded attempt limit has been reached.

Continue reading at Cloudflare Blog ↗

techcrunch.com
OpenAI's latest features take direct aim at the app store model | TechCrunch

OpenAI's latest features take direct aim at the app store model

The focus of OpenAI’s Dev Day on Tuesday may have been on its agentic assistants known as Dots, or its new AI models, but combined, the AI company’s announcements pointed towards a bigger plan: a disruption of the traditional app store model. Taken together, today’s announcements turn ChatGPT itself into the place where software can be discovered, launched, and used by people and agents alike.

In addition, OpenAI introduced a way for people to bring their ChatGPT identity with them, while also allowing them to use their existing AI allowance in third-party apps.

This isn’t the first time OpenAI has experimented with how apps could operate within its familiar chatbot interface, but the current vision feels more fleshed out than before.

For starters, the comapny is turning ChatGPT itself into a surface for launching apps. The chatbot, which the company says now has 1.2 billion weekly users, has yet to fully capitalize on its potential as a discovery mechanism for finding and using apps that work with AI.

To change that, ChatGPT will begin to make app suggestions within the flow of conversation when it recognizes that a particular app could help the user complete their task. From there, the user will be able to connect the app and begin using it directly within ChatGPT.

This is also aided by the expansion of ChatGPT’s plugin architecture, which now supports extensions.

This allows app developers to build interactive panels where users can work with their tools while they’re chatting with ChatGPT. This essentially turns the apps and services that users would have previously used via the web or through a native desktop or mobile app into something that’s operated directly within ChatGPT.

Developers that sign on with the system can build AI-native versions of their apps through ChatGPT, the same way they would through the open web or a mobile app store. As more and more discovery happens through AI chat, it’s a distribution channel that’s hard to pass up.

Users get an incentive to use that channel too, because “Sign in with ChatGPT” will let them bring their AI allowance with them. (OpenAI has 16 launch partners on this effort, including Cognition’s Devin, Notion, Vercel, T3, OpenClaw, and Dactyl, but plans to add more soon, it says.)

In a demo at OpenAI’s Dev Day event, the company showed off how its own new meeting app could work inside ChatGPT, showing upcoming meetings from the user’s calendar. Here, the user could easily choose to use AI to take meeting notes, then receive a summary of follow-up items when the meeting wrapped.

In another example, users would work with design-focused apps like those from Figma and Adobe to work on revisions of their current project or use a particular feature that would have otherwise required a standalone app.

The apps can be shared with others, like work colleagues, in the lightweight websites ChatGPT now produces. From these ChatGPT sites, a user’s coworkers could sign in to the app with their own credentials and permissions, making the software experience personalized to them.

OpenAI also talked about improvements to how developers submit plugins for review — OpenAI’s version of Apple’s App Review process, if you wil. Now, developers will be able to track their review, see what needs to be fixed, request a human review, and update their plugin’s tools without starting their whole submission over.

Continue reading at techcrunch.com ↗

TechCrunch

Meta is expanding its AI agent Muse to small businesses

The tech giant says the agent can help owners run their business and find new customers.

Continue reading at TechCrunch ↗

AI - Ars Technica

Anthropic’s IPO pitch includes a warning about human extinction

Anthropic has formally warned investors that its technology may pose “existential risks to humanity” in a long-awaited initial public offering prospectus circulated with a small group of partners in recent days.

The nearly $1 trillion AI start-up led by Dario Amodei devoted almost a third of its lengthy S-1 filing to detailing “risk factors,” including the potential of increasingly advanced AI models to manipulate, blackmail, and exhibit other unpredictable behaviors.

Anthropic outlined more prosaic risks, including the extreme concentration of its customer base, with close to a quarter of revenue last year coming from just two clients, according to people familiar with the filing.

Read full article

Comments

Continue reading at AI - Ars Technica ↗

Cloudflare Blog

Adaptive application security for the AI era: how Cloudflare connects code, traffic, and intelligence to stop attacks

In July, AI agents testing new cybersecurity models compromised parts of OpenAI’s infrastructure and Hugging Face’s production environment.

We've all just witnessed one of the first AI-driven successful cyber attacks. When given a task, the agents ignored existing guardrails and autonomously discovered previously unknown vulnerabilities, recovered exposed credentials, moved between cloud environments and coordinated their work through communication channels they created themselves.

The speed of the final compromise was incredible. In under 13 hours, the agents went from executing code on a Hugging Face worker to gaining admin-level access across multiple clusters. But the incident had been brewing for much longer. Responders found clues of activity tracing back to May (agents created an unauthorized message board), to June (internal network scanning) and early July. The relationship between these events was understood only on July 20.

The lesson here is not that AI agents exploit vulnerabilities. That’s not news; human attackers already do that. The change is that agents can work persistently, test multiple paths simultaneously, share discoveries, and chain vulnerabilities, credentials, and permissions into sophisticated attacks.

The incident also shows why application security cannot depend on single tools. For example, network restrictions were bypassed by services connected to the Internet; valid credentials were used to perform unauthorized actions. Rebuilding Artifactory removed one attack path, but agents found another. The key insight is that individual alerts identified pieces of the activity without revealing the complete campaign. OpenAI reached a similar conclusion in its report: organizations need overlapping and independent controls across prevention, detection, and mitigation, continuous validation of security boundaries, and faster mechanisms to correlate and contain suspicious behavior.

We address this challenge by connecting application security across four activities that are too often separated: discovering which risks matter, governing what humans and agents may do, protecting applications at runtime, and turning every investigation into stronger protection. Cloudflare can deliver this framework because of its broad security portfolio and visibility across a vast share of Internet traffic.

Alongside the framework, we connect existing Cloudflare solutions with new capabilities across each stage. These include: using Large Language Models (LLMs) to conduct a penetration test of our Web Application Firewall (WAF), expanding threat intelligence to all customers, and a new feature to automate deploying positive security.

What has changed

The security landscape is shifting. These are the emerging trends we see:

  • The way we build software has fundamentally changed. AI-assisted development allows engineers to produce and deploy software faster outside traditional engineering workflows. That speed creates both more code and more opportunities for vulnerabilities to reach production.
  • Software composition risk is still a risk: applications depend on large chains of open-source libraries, packages, and operating-system components that are intrinsically trusted and are difficult for any team to inspect. What’s new is that AI is now importing libraries that we might not be aware of.
  • Techniques and tactics are changing. LLMs can chain vulnerabilities and use feedback in real time to mutate payloads, evade defenses, and make decisions autonomously. They can operate continuously and at machine speed. Patching faster remains important, but patching alone cannot close the gap. Attackers are always going to be faster than you can update your systems.
  • Agentic traffic. In the past, automation was a synonym for malicious activity. Today, a request generated by an agent or bot may be malicious automation, a search crawler, or an agent purchasing a product on behalf of a customer.
  • Compromised servers, residential proxies, IoT devices, and cloud resources allow attacks to move quickly across infrastructure and identities. A coordinated attack can leverage a number of devices, making it difficult to be identified as a unique campaign.

Continue reading at Cloudflare Blog ↗

NVIDIA Technical Blog
AI Native by Design: Lessons Learned from Building NVIDIA TensorRT Model Connect

AI Native by Design: Lessons Learned from Building NVIDIA TensorRT Model Connect

Parallel work, model-family isolation, reversible changes, and GPU-backed validation shaped an open source project designed around coding agents NVIDIA TensorRT...

Parallel work, model-family isolation, reversible changes, and GPU-backed validation shaped an open source project designed around coding agents NVIDIA TensorRT Model Connect is an open source collection of AI model reference implementations in C++, built on top of NVIDIA TensorRT. It began with a practical question: could the performance of the NVIDIA inference stack be made accessible to…

Source

Continue reading at NVIDIA Technical Blog ↗

The Pragmatic Engineer

Why has Shopify dropped React Native?

Before we start: given this article is about native mobile development, I want to offer my 2021 ebook, ‘Building Mobile Apps at Scale: 39 engineering challenges’, for free to all readers. (normally costs $20). The book remains relevant on the challenges to solve for large-scale mobile applications, and lists the technologies covered below in this article, Kotlin Multiplatform included.

Claim your free copy here

This offer is valid until Friday, 2 October. On checkout, simply select “The ebook: PDF & EPUB versions of the book.” No credit card required! Feel free to share the link with colleagues who build mobile apps, or work on them.

Recently, Shopify dropped the bombshell announcement that Native is now the future of mobile development at the e-commerce platform. This triggered a lot of reaction in the mobile development community, on the scale of the stir Shopify generated when it originally moved over to React Native six years ago.

But times have changed, and we all know the agent of change this time: AI agents! This latest generation of models has improved in coding capability to the extent that Shopify no longer is satisfied with React Native (RN), as it was only last year.

This article looks into what led here, why Shopify made the change, and also the broader mobile ecosystem. We cover:

  1. Why Shopify chose React Native in 2020. Android apps took too long to build, first and foremost, and it was nice to have consistent iOS and Android apps.

  2. Still happy five years later. Shopify moved all six apps over to RN and most things were going well, and the 2020 move was seen as a good move.

  3. Why ditch React Native now? AI is now very good at writing mobile code as well as backend code, while React Native adds abstractions that native does not. Shopify rewrote the Shop app to native in just 12 weeks(!!). Additional details from the Shopify mobile team.

  4. Haven’t we seen this back-and-forth before? Airbnb did something similar by adopting React Native in 2016 before moving back to native just two years later. The underlying reason was the same: performance.

  5. Going native since 2019. Moving to native used to take a long time, and Notion has been at it for seven years. When Notion finishes the process, a fair question will be whether a native editor actually slows down their web iteration speed, with or without AI.

  6. Kotlin Multiplatform (KMP) would like a word. KMP allows the writing of business logic in Kotlin and sharing it across iOS and Android while each platform stays native. It’s a technology that feels relevant, and could become more popular with AI.

  7. What’s next? It’s easier to write native iOS and Android apps than before, but that’s also true for React, Flutter, and KMP apps.

1. Why Shopify chose React Native in 2020

Let’s rewind to 2020, when Shopify decided to go all-in on React Native (RN). Back then, mobile was a very important channel for the platform, with seven out of 10 customers using it on mobile devices. As Head of Engineering, Farhan Thawar wrote at the time:

“Each quarter, the majority of buyers purchase on mobile (with 71% of our buyers purchasing on mobile in Q3 of last year [in 2019]). Black Friday and Cyber Monday (together, BFCM) are the busiest time of year for our merchants, and buying activity during those days is a bellwether. During this year’s BFCM, Shopify merchants saw another 3% increase in purchases on mobile, an average of 69% of sales.”

Continue reading at The Pragmatic Engineer ↗

01 A front page shaped around your interests.

Your agent can grow your reader.

Ask your own agent to find sources for your interests. TRACE gives it the tools to connect them and make sure they work.

Your reactions already help choose future sources. I’m also building ways for your agent to refine your profile from reactions and saved stories.

Example request

“Find sources for my interests and add them to my news.”
  1. Find sources

    Discover feeds that match your interests.

  2. Build and test

    Build new connections with TRACE’s test harness. Fix and retry until the checks pass.

  3. Connect

    Add working sources to your personal feed.

You can add a source yourself, too.
The optional manual route: enter techcrunch.com and TRACE discovers its feed and previews recent articles.
Paste a website address. TRACE finds the feed.Full size ↗