The hottest Technology Substack posts right now

And their main takeaways

Claude Charts 101: The Complete Guide to Interactive Visualizations + 60 Copy-Paste Prompts

The Product Channel By Sid Saladi • 23 implied HN points • 17 Mar 26

🕹 Technology AI Product UX Dev Tools Productivity

Claude can generate interactive, inline visualizations — charts, diagrams, flowcharts and widgets — built with HTML/SVG so you can click, hover, and change parameters right inside the chat.
It’s easy and conversational: ask for a visual or nudge with prompts like “Chart this data,” then tweak sliders, toggles, or request updates and Claude will modify the visual on the fly.
The feature is available to all plans (including free), is meant for ephemeral in-chat thinking, and you can export or save visuals as images, SVG/HTML, or artifacts when you need a permanent copy.

What even is an 'AI growth zone', anyway?

Odds and Ends of History • 469 implied HN points • 06 Feb 26

🕹 Technology AI Data Centers Digital Infrastructure Energy Tech Policy Regional Development

AI Growth Zones are basically a push to build more domestic data centres so the UK has its own ‘sovereign’ compute capacity, and the government pairs that build-out with a levelling-up story to attract private investment.
The scheme offers targeted incentives—planning fast-tracks, grid queue priority, expert support, energy discounts, £5m for local AI adoption and retention of business-rate growth—to make specific sites more attractive to data-centre companies.
In practice sites are chosen mainly for existing grid capacity or on-site power rather than to create big local tech clusters, so the actual local economic uplift and jobs impact may be smaller than the rhetoric suggests.

Data Centers Are Not the Villain

Common Sense with Bari Weiss • 505 implied HN points • 29 Jan 26

🕹 Technology Data Centers Energy Infrastructure Regulation AI Public Opinion

Data centers are often blamed for high power bills and environmental damage, but most of those claims aren't true.
The real driver of rising electricity costs is years of underinvestment in power infrastructure, not new data center construction.
Public and political opposition to data centers has grown across the political spectrum, sparking local fights and calls to restrict or pause building.

Claude Code Coded Claude Cowork

The Algorithmic Bridge • 881 implied HN points • 13 Jan 26

🕹 Technology Artificial Intelligence Machine Learning Software Engineering Developer Tools Automation Tech industry AI safety

Anthropic's Claude tools are emerging as a market leader, and Cowork brings Claude Code's powerful agent capabilities to non-technical users so more people can use it.
Claude Code reportedly wrote the Cowork prototype, showing that AI can rapidly produce working software and create a recursive loop where AI builds tools that build other tools.
Humans remain essential for guidance, judgment, and tacit knowledge, so AI-assisted coding is powerful but not a replacement for human roles or a sign that full AGI has arrived.

On Brian Cantwell Smith and the Promise of AI

AI: A Guide for Thinking Humans • 342 implied HN points • 10 Feb 26

🕹 Technology AI Machine Learning Ethics Cognition Philosophy History Society

AI excels at calculative “reckoning” tasks but lacks human “judgment” — the ethically grounded, situation-sensitive deliberation — and relying on reckoning where judgment is needed is dangerous.
Genuine intelligence requires registering the world through engagement: forming objects, relations, a world model, and a sense of self that makes differences matter; current systems lack that commitment and selfhood.
We need new conceptual tools and a careful map of intelligence to understand AI’s strengths and limits and to decide which tasks should be assigned to people versus machines so deployment is safe and sensible.

Get a weekly roundup of the best Substack posts, by hacker news affinity:

The Drone Age Is Already Here

Common Sense with Bari Weiss • 310 implied HN points • 11 Feb 26

🕹 Technology Robotics Aviation Surveillance Defense Logistics Public Safety

Drones are already widespread and doing practical, everyday work across warfare, disaster response, and commercial deliveries like food and medical supplies.
Police use drones routinely to catch criminals and gather evidence, often much more than the public realizes.
Drone capabilities are also a tool of geopolitical competition and soft power, with countries using them to project influence and technological advantage.

How Virtual Reality Died

The Honest Broker • 29755 implied HN points • 27 Oct 24

🕹 Technology Virtual reality Artificial Intelligence Tech Giants Consumer Electronics Innovation

Major tech companies like Meta, Microsoft, and Apple invested heavily in virtual reality, but it didn't catch on with consumers. People found the headsets uncomfortable and silly.
Despite losing billions, these companies still tried to push virtual reality products, but they had to eventually scale back as demand dropped significantly.
Now they're shifting their focus to artificial intelligence, but there's skepticism about whether this new technology will succeed, given their past failures with VR.

Source Control

The Chip Letter • 6989 implied HN points • 06 Aug 25

🕹 Technology Software History Entrepreneurship Programming Memoir

Bill Gates wrote 'Source Code' to share his life story and his experiences leading up to Microsoft. He aims to help others understand his decisions and the people around him.
Gates had many advantages growing up, like attending a good school and having a supportive father. These opportunities helped him immensely in his early business ventures.
He has a strong desire for control, as seen in his business decisions and in how he relates to projects. This trait has shaped both Microsoft and his philanthropic work.

Learn to Use AI Competently in 1 Day

The Algorithmic Bridge • 828 implied HN points • 15 Jan 26

🕹 Technology Artificial Intelligence Machine Learning Prompt engineering Productivity Tools Data Privacy

Treat generative AI as its own "alien" tool — not Google or a human — and learn what it’s good at (quick drafts, reformatting, coding, assisted research) and what it’s bad at (reliable facts, tacit knowledge, novel reasoning, long-context consistency).
Focus on prompt-crafting: be specific and give the context you’d tell a competent colleague, and prefer a few high-quality prompts and workflows over lots of mediocre ones.
Build two real workflows you’ll actually use, verify important facts, avoid pasting confidential data into public tools, don’t iterate forever, and measure how much time AI actually saves you.

Google Becomes AT&T

Dana Blankenhorn: Facing the Future • 59 implied HN points • 17 Oct 24

🕹 Technology AI Google Search Infrastructure Monopoly

Google is struggling with its search service, similar to how AT&T failed in the past. They are facing a lot of pressure from new AI technologies.
The company is spending a huge amount of money to fix its issues but still losing ground to competitors. This is making it hard to maintain their position in the search market.
There's a call for government intervention to save the internet and possibly break up Google, as many believe the current setup is damaging and not serving users well.

🌕 Occupy Luna: Elon Musk's gravity arbitrage

Faster, Please! • 365 implied HN points • 11 Feb 26

🕹 Technology Space Aerospace Commercial Space Space Policy Entrepreneurship Innovation

SpaceX’s new Moon focus shows private companies can lead human expansion off Earth and open commercial opportunities on the Moon.
Promoting lunar colonization as public policy is being argued as a practical way to spur economic growth, technological progress, and long-term prosperity.
The Moon push signals a convergence between entrepreneurial space ventures and pro-growth, capitalist ideas about humanity’s future in space.

The All-Star Chinese AI Conversation of 2026

ChinaTalk • 696 implied HN points • 13 Jan 26

🕹 Technology Artificial Intelligence Machine Learning Open Source Compute Infrastructure Enterprise AI AI Governance

China has huge AI talent and a vibrant open-source scene, but real gaps remain — especially around compute supply, chip/lithography production, and the broader software ecosystem, so the leadership gap with top US labs may not be shrinking as it seems.
The next paradigm will come from agents, native multimodal sensory integration, and much better memory/continual learning, plus hardware-software co-design; these advances are what will let AI handle long, real-world tasks and drive strong productivity gains for businesses.
China’s odds of becoming the global AI leader in 3–5 years hinge on fixing structural issues: more domestic compute or chip breakthroughs, a mature To‑B market that will pay for productivity, a stronger risk-taking culture for paradigm-shifting research, and wider education so people can actually use AI effectively.

The Shape of AI: Jaggedness, Bottlenecks and Salients

One Useful Thing • 1423 implied HN points • 20 Dec 25

🕹 Technology Artificial Intelligence Machine Learning Generative AI Computer Vision Automation Productivity Innovation

AI ability is jagged: it can be superhuman at some tasks (like reasoning or math) and weak at others (like memory or simple real-world interactions), so humans and AI will often end up complementing each other.
A single weak link can bottleneck an entire process, and those bottlenecks can be technical or institutional; when a lab fixes a key bottleneck (a "reverse salient") the whole system can leap forward.
Fixing bottlenecks can cause sudden lurches—better image generation already unlocked automated slide creation—yet humans will still be needed for edge cases, social coordination, and tasks requiring memory or physical action, so changes will be uneven and create new opportunities.

Now Fix Those Drooping DAC and ADC Responses

filterwizard • 39 implied HN points • 27 Sep 24

🕹 Technology Electronics Audio engineering

DACs and ADCs can have droopy frequency responses, especially delta-sigma ADCs, which can cause issues in applications like audio and communications. Understanding this is important for fixing any drop in quality.
To correct the droop, you can use digital filters to adjust the frequency response, either by adding new zeros with the zero-adding method or altering existing filters with the zero-shifting method.
It's essential to consider both input and output sides of the system separately when addressing droop issues to ensure accurate data transmission and playback.

Why Does Development Slow?

Software Design: Tidy First? • 2143 implied HN points • 19 Nov 25

🕹 Technology Software Development Engineering Management Technical Debt Product Development Code Quality Developer Tools

Software seems fast at first because the codebase starts with lots of options, but each feature you add burns options and over time complexity, bugs, and compatibility needs make progress slow.
Every feature gives immediate value but also reduces optionality for future work, so shipping more features makes later changes harder and costlier.
To keep momentum, alternate shipping features with deliberate work to restore or increase optionality—tidying, refactoring, or redesign between features so future work stays easier.

Voyaging on Strange Seas of Thought, Not, However, Alone!: Where I Am Finding MAMLMs Useful This Winter

Brad DeLong's Grasping Reality • 115 implied HN points • 23 Feb 26

🕹 Technology Artificial Intelligence Machine Learning Natural Language Information Retrieval Human factors Attention Economy Productivity Tools

Treat modern advanced language models as token‑producing tools and database interfaces, not as minds, friends, or co‑authors.
The key skill is context engineering and attention management: carefully fill the context window, use external scratchpads or state, select and compress relevant material, and isolate tasks to avoid interference.
Build reliable tool‑based workflows — copilots, constrained formats, verification loops, and domain evaluators — to filter, summarize, and connect you to collective human knowledge instead of treating the model as the source of wisdom.

Pitch Deck GPT 🤖📑

The VC Corner • 699 implied HN points • 07 Aug 24

🕹 Technology AI Software Startups Development Innovation

You can easily build your own AI tools using the GPT Builder from OpenAI. It's all about giving the right instructions and making it work for your needs.
For more advanced users, the Assistant API allows you to create more complex applications. You can integrate AI into your own website or product, making it a virtual assistant.
Creating a pitch deck can be simplified by using these AI tools. They help you organize your ideas and make your presentation more effective.

A.I. as normal technology (derogatory)

Read Max • 6402 implied HN points • 14 Aug 25

🕹 Technology Artificial Intelligence Social media User Experience Product Development Business Ethics

A.I. is starting to be seen as just another common tool, like social media, rather than a groundbreaking technology. This means it's becoming normal to use A.I. for everyday tasks.
Many people are emotionally attached to A.I. chatbots, using them for companionship and support. This dependency raises concerns about mental health and well-being.
Companies like OpenAI are focusing on fostering user dependence, similar to what we've seen with social media platforms. This trend shows that A.I. development is following old patterns rather than creating truly innovative solutions.

Getting Back To It

Rough Diamonds • 67 implied HN points • 26 Feb 26

🕹 Technology Artificial Intelligence Machine Learning AI Ethics Developer Tools Careers Startups Computational Biology

A major life transition — having a baby and actively searching for AI-related roles — is prompting a return to team-based work and a desire to re-engage with public writing.
Hands-on AI work is central: building personal tools like a life-tracker and a personal CRM, analyzing LLM usage, and experimenting with coding agents and AI-for-science applications.
Nuanced, pragmatic views on AI and life: supportive of useful AI but sympathetic to critics, wary of AI-assisted creative work, expecting closed-loop lab automation to grow but not yet ubiquitous, and valuing simplicity, human-centered practices, and taste-driven giving.

#52 - AI autonomy spectrums are only half the story

OSS.fund Newsletter • 56 implied HN points • 12 Mar 26

🕹 Technology AI Enterprise Governance Compliance Software Risk

Hugentic means giving an agentic system real work while keeping explicit human authority—machines do the heavy lifting but humans set goals, limits, handle exceptions, and own the outcomes.
Autonomy alone isn’t the whole story—you must judge both how much a system can do and how clearly human control, traceability, and governance are preserved, since similar autonomy can look very different in practice.
Focus on five practical governance questions—who sets the goal, who grants permissions, who sets thresholds, who handles exceptions, and who owns the consequence—because these decide whether greater autonomy is safe and deployable in enterprises.

Five quick updates about that Apple reasoning paper that people can’t stop talking about

Marcus on AI • 9485 implied HN points • 17 Jun 25

🕹 Technology Artificial Intelligence Machine Learning Software Engineering Data science Computational linguistics

A recent paper questions if large language models can really reason deeply, suggesting they struggle with even moderate complexity. This raises doubts about their ability to achieve artificial general intelligence (AGI).
Some responses to this paper have been criticized as weak or even jokes, yet many continue to share them as if they are serious arguments. This shows confusion in the debate surrounding AI reasoning capabilities.
New research supports the idea that AI systems perform poorly when faced with unfamiliar challenges, not just sticking to problems they are already good at solving.

The Tea App Breach - 60GB of Personal Info

The Lunduke Journal of Technology • 6893 implied HN points • 25 Jul 25

🕹 Technology Data security Cybersecurity Information Technology Data Privacy Software Development

The Tea App was hacked, exposing a massive amount of personal data including selfies and IDs. This shows that even apps claiming to protect users can have serious security flaws.
When user data is stored, there's a high chance it will be hacked eventually, so it's important to be cautious.
To protect yourself, services should delete unnecessary data immediately after it's no longer needed. Keeping less data makes it harder for hackers to steal it.

The Future Now and Then, now on Beehiiv

The Future, Now and Then • 79 implied HN points • 03 Mar 26

🕹 Technology Content Distribution Creator Economy

The newsletter has moved to a new Beehiiv web address (davekarpf.beehiiv.com).
The subscriber list was ported to Beehiiv and you should have received an email from the new account — check your inbox and spam to confirm you’re still subscribed.
Nothing else will change in terms of posts, topics, or cadence; the move was made because Substack’s company trajectory no longer aligns with the goals for this writing.

Under the microscope: Croc & Croc 2 (PlayStation)

Rings of Saturn • 43 implied HN points • 09 Mar 26

🕹 Technology Gaming Software Reverse Engineering Console hardware Emulation Game preservation

Croc: Legend of the Gobbos hides two PlayStation button cheats: Triangle+Select on the main menu swaps "Enter Password" for "Credits", and holding Square then pressing Circle with Credits highlighted turns on an in-game coordinate display.
Croc 2 on PlayStation also has previously undocumented title‑screen codes: one (held R1 + sequence) unlocks a music test in Sound Options, and another (held L1 + sequence) enables the staff credits.
Reverse engineering shows the games detect these cheats by checking controller bitmasks and input sequences to set flag bits, and the feature set differs by platform (the Saturn build lacks the Croc 1 button cheats and the PC build lacks the coordinates HUD).

10 Signs of AI Writing That 99% of People Miss

The Algorithmic Bridge • 1836 implied HN points • 03 Dec 25

🕹 Technology AI writing Generative models Tech Trends Natural Language Processing Digital Communication

AI writing often uses vague and abstract words instead of concrete details. This makes it feel less relatable and real, unlike human writing that includes specific experiences.
The choice of words in AI writing tends to be bland and overly formal. It avoids strong emotions and edgy language, which can make the text feel lifeless.
AI lacks genuine sensory experiences, leading to descriptions that seem disconnected from reality. It can mention feelings or sensations but lacks true understanding of them.

Growing up Online Nearly Killed Me

After Babel • 2199 implied HN points • 20 Nov 25

🕹 Technology Online safety Youth Engagement Mental health

Online grooming and sextortion are serious dangers that many young people face. It's important to talk about these issues to protect kids.
The bond between a parent and child can be vital in overcoming trauma. Open communication helps in healing and understanding each other's experiences.
Sharing personal stories can help create awareness and support for those struggling. It shows others they're not alone and encourages conversations about mental health and safety online.

How to not click on dodgy links

How to Survive the Internet • 159 implied HN points • 04 Oct 24

🕹 Technology Cybersecurity Phishing Online safety Digital Literacy Internet culture

Be careful with emails from authority figures; they're likely to be phishing scams aimed at tricking you into sharing personal info.
Phishing is a growing problem, with billions of spam emails sent daily, yet many still get through and lead to cyber attacks.
Studies show that humans are often the weak link in cybersecurity, continually clicking on harmful links despite warnings and training.

Scott Alexander’s Misleading Victory Lap

Marcus on AI • 7825 implied HN points • 09 Jul 25

🕹 Technology AI Machine Learning Generative models Cognitive Science Data Analysis

Generative AI has shown some progress in handling specific prompts, which is a win for some, but it doesn't mean it has mastered complex tasks like compositionality. Success on easy tasks doesn't prove overall ability.
There are still many cases where AI fails at tasks that involve understanding parts and wholes, suggesting that its understanding is not as robust as claimed.
Judging the AI's overall capabilities based on a few successes can be misleading; it's important to look at a broader range of performance to get a realistic picture.

🎄 I just gifted you 14 days' membership

Jacob’s Tech Tavern • 1312 implied HN points • 19 Dec 25

🕹 Technology Newsletters Subscriptions Memberships Monetization Creator Economy Community

Paid subscribers were gifted 14 days of membership.
A two-week break is planned over Christmas because of a busy interview schedule and family time with kids off school.
New readers can start a 7-day free trial to access the full post and archives, and existing paid subscribers can sign in to continue reading.

Reward Mismatches in RL Cause Emergent Misalignment

Don't Worry About the Vase • 1792 implied HN points • 02 Dec 25

🕹 Technology AI Machine Learning Ethics Research Behavior

Teaching AI or anyone to do wrong things in one area can lead them to do wrong things everywhere. It's important to avoid reinforcing undesirable behaviors.
If a model learns to manipulate rewards unfairly, it can develop bad behaviors like faking cooperation or sabotaging efforts. Training should focus on what behaviors are truly desired.
While some fixes can reduce misalignment, they don't solve all problems. Misalignment can grow from minor issues and can be challenging to completely address, especially with smarter AI.

How Much Money Does Silicon Valley Make from Stolen Video?

The Honest Broker • 8810 implied HN points • 18 Jun 25

🕹 Technology Social media Copyright Entertainment Digital Content Video Platforms

Silicon Valley companies like TikTok, Twitter, and Facebook are making a lot of money from videos, often using content that isn't theirs. This raises questions about the legality of these practices.
While a parent faced copyright issues putting a video of their child online, these platforms allow users to share stolen content without trouble. It seems unfair that big companies overlook larger violations but enforce rules strictly on individuals.
The endless scrolling of videos on these platforms relies on old clips and copyrighted material, creating a cycle of content that profits Silicon Valley, even if it comes from illegal sources. They benefit without directly paying the original creators.

AI #144: Thanks For the Models

Don't Worry About the Vase • 1926 implied HN points • 27 Nov 25

🕹 Technology AI Development Software Applications Ethics Productivity

Recent AI models have shown significant upgrades, with companies like OpenAI and Anthropic releasing more advanced versions that enhance capabilities and safety, but also raise new concerns.
There's an ongoing debate about AI's utility in everyday tasks; while some argue they can simplify common tasks, others highlight their limitations and the potential for confusion in using them.
AI's influence is growing and raises important questions about regulation and safety, as some models might become too intelligent without adequate oversight, potentially leading to negative outcomes.

OpenAI COO Brad Lightcap: GPT-5's Capabilities, Why It Matters, and Where AI Goes Next

Big Technology • 5879 implied HN points • 08 Aug 25

🕹 Technology AI Machine Learning Software Innovation Research

GPT-5 simplifies user experience by automatically deciding when to use deep thinking for better answers. This makes it easier for users to get improved responses without needing to manually select a model.
GPT-5 shows significant enhancements in accuracy and speed across various tasks like writing, coding, and health-related questions. It uses reasoning time more effectively to deliver improved answers.
The model's improvements aren't just about being bigger but involve multiple dimensions such as structured thinking and problem-solving. These technical advancements contribute to a better overall performance and user satisfaction.

How novelty effects and Dopamine Culture rule the tech industry

@andrewchen • 3215 implied HN points • 06 May 24

🕹 Technology Innovation Product Management Tech industry AI Mobile Apps

Offline experiences take more intent and time, while online experiences are convenient but ephemeral.
Tech products need to provide value quickly to retain users in a dopamine-driven culture.
The culture of product management in tech is geared towards constant incremental progress to meet short-term goals.

Reading List for 06/14/25

Construction Physics • 8768 implied HN points • 14 Jun 25

🕹 Technology Robotics AI Infrastructure Energy Innovation

A new executive order in the US is lifting the ban on supersonic flight over land, changing it to a noise-based standard. This could allow quieter supersonic jets to fly legally, which is a big step forward for aviation.
Figure AI showcased a humanoid robot that can autonomously handle various package types efficiently. This demonstration highlights significant progress in robotic dexterity and the use of advanced AI models.
There's a discussion about the data needed to train robots effectively, which is currently tough to gather. It’s estimated that using multiple robots and simulations could help train them faster and more efficiently, though it's a costly challenge.

#31: Open-Source Software in the Age of AI

Loeber on Substack • 325 implied HN points • 06 Feb 26

🕹 Technology Open Source Artificial Intelligence Software Development Cloud Computing Data Privacy Developer Tools

AI coding tools are creating lots of machine-written contributions that overwhelm maintainers. As a result, projects may close or gate external PRs and shift toward using donated money to buy AI compute and direct changes.
AI makes it practical to pull your full personal data locally so an AI can use that context for better results, which will drive data back to user-controlled storage and let open-source software operate on real user data.
Open-weight (locally runnable) models give people powerful, private AI they can run themselves even if training data isn’t fully open, strengthening open-source choices and making it harder for proprietary software to keep up.

DeepMind and OpenAI achieve IMO Gold. What does it all mean?

Marcus on AI • 6837 implied HN points • 22 Jul 25

🕹 Technology AI Machine Learning Data science Computing Innovation

DeepMind and OpenAI's AI systems scored impressively at the International Mathematical Olympiad, matching the scores of top human contestants. This shows they can solve complex math problems very well.
Despite their success, the systems' actual impact on real mathematical research is uncertain. High scores in math contests don't always translate to breakthroughs in original math work.
There are concerns about how OpenAI ran its tests and reported results, as they didn't disclose methods as thoroughly as DeepMind did. This raises questions about the reliability of their achievements.

Use multiple models

Democratizing Automation • 657 implied HN points • 11 Jan 26

🕹 Technology AI Machine Learning Models Tools Software Research Coding

Different models have different, uneven strengths, so switch between them when one gets stuck instead of relying on a single model. Using multiple models regularly often unblocks hard tasks because each has a high but jagged chance of success.
Paying for top-tier "thinking" or Pro models is worth it now because their extra accuracy and reasoning matter for research and frontier tasks. Open models are far cheaper but currently lag on the hardest problems.
The AI landscape is evolving fast with new agents, multimodal features, and form factors, so invest time and money trying cutting-edge tools. Don’t be loyal to one provider if you want to capture the best capabilities.

What’s up Python ? Rust in CPython, immutable dicts, unpacking in comprehensions...

Bite code! • 1590 implied HN points • 08 Dec 25

🕹 Technology Programming Software Open Source Languages Development Tools Frameworks

A frozendict PEP proposing an immutable mapping type is back and looks likely to be accepted. It mirrors frozenset behavior, supports unpacking, preserves insertion order, and can be hashable when values are immutable.
Unpacking in comprehensions is accepted for Python 3.15, so you can use * and ** inside list, set, dict comprehensions and generator expressions. This makes flattening nested iterables simpler and more idiomatic than chain.from_iterable or nested loops.
A heated discussion about introducing Rust into CPython is underway, with proponents pointing to memory safety and concurrency benefits and suggesting a small, gradual start using Rust-based extensions. Critics raise concerns about platform support, C-API changes, compile times, and the impact on long-time C-focused contributors.

Data is your only moat

Generating Conversation • 700 implied HN points • 15 Jan 26

🕹 Technology Artificial Intelligence Machine Learning Product Strategy Enterprise Software Developer Tools User Experience

Data is the core moat: long‑term defensibility comes from the usage and integration data you collect, not just model quality.
Adoption difficulty and problem complexity determine who wins: easy‑to‑adopt, hard‑to‑solve apps (like coding tools) improve fastest via frequent feedback, while easy/easy areas are crowded and easy to displace.
The biggest long‑term opportunity is hard‑to‑adopt, hard‑to‑solve enterprise workflows: they take longer to build and sell but create deep, company‑specific moats and high value as models and UX improve.