103 AI use cases that were impossible two years ago:
Marketing
1. Ad graveyard analysis. Every morning, the agent pulls competitors’ active ads from the Meta and LinkedIn ad libraries and logs when each one appears and disappears. Ads that run for months are winning, and ads killed within a week failed. You get their winners ranked, with copy and creative.
2. Short-lived account pages. When a target account shows buying intent, like repeat pricing-page visits or a demo request, the agent builds a landing page just for them. It includes their logo, use cases for their industry and relevant case studies. The agent sends the link to the account owner and takes the page down after 30 days.
3. AI-answer share tracking. Every night, the agent asks ChatGPT, Claude and Perplexity 500 buyer questions and logs whether you’re mentioned, where you rank, and which sources were cited. Each week, it lists the questions where competitors show up and you don’t. It drafts content to close each gap.
4. Outage ads. The agent checks competitors’ status pages every few minutes. When one reports a major outage, it switches on a pre-approved “having trouble today?” campaign on their brand terms, and turns it off when service is restored.
5. Ideal-customer engagement alerts. The agent pulls everyone who opens your newsletter or engages with your posts on X and LinkedIn, and looks up their company, title and company size. When someone matches your ideal customer profile, it alerts you with who they are, what they engaged with, and a suggested opener.
6. Free-text attribution. The agent reads every “How did you hear about us?” answer from forms and sales calls, cleans up messy answers, and groups them by source. Each month, you get the podcasts, Slack groups and people driving pipeline that analytics can’t see.
7. Chatbot transcript mining. Each week, the agent reads every chatbot conversation, groups the questions, and checks whether your site already answers each one. You get the copy gaps ranked by frequency, with suggested copy.
8. Handwritten notes on engagement. When someone who fits your ideal customer profile engages with your content, the agent finds their work address and writes a note referencing the specific post. It sends the note through a robot-handwriting service, so a real pen-written card lands on their desk.
9. Spotify Wrapped for every customer. Each December, the agent pulls each account’s usage and outcomes for the year, and calculates headline stats like hours saved and biggest wins. It designs a branded year-in-review for every account that champions will forward internally.
10. Personal answers to webinar questions. The agent logs every question in the webinar chat and who asked it. For each one that didn’t get answered live, it drafts an answer from your docs and the speaker’s points, and emails it to that person the next morning.
Content
11. “I called this” reposts. When news breaks in your space, the agent searches everything you’ve ever posted, finds the post where you predicted it, and drafts a repost linking the news.
12. Format switching. Each week, the agent tracks which formats are taking off on each platform. It re-drafts your best-performing ideas in those formats.
13. The first 30 minutes of a comment. The agent watches posts from executives at your target accounts. When one posts on your topic, it alerts you within minutes with a drafted comment, so you post while the thread is still small.
14. Inbox-to-post mining. The agent scans your sent emails and DMs for questions you’ve answered three or more times. It drafts a post from your best answer, because the demand is already proven.
15. Where viewers leave your videos. The agent matches every sharp drop in YouTube audience retention to what you were saying at that moment. Across all your videos, it finds the patterns that lose people and flags them in new drafts.
16. Unanswered YouTube searches. The agent pulls YouTube autocomplete suggestions in your niche and checks the top results for each. It flags searches with no good video yet, with a suggested title and outline.
Competitive intelligence and trends
17. Filings as early signals. Every week, the agent checks trademark, patent and FCC databases for competitors’ new filings. It summarizes what each one reveals, like a new product name or device, months before launch.
18. Developer traction. The agent tracks GitHub stars, npm downloads and Hugging Face downloads daily for competitors and open-source tools in your space. It alerts you when one starts growing unusually fast, before the press notices.
19. Earnings-call language. The agent reads every earnings call transcript in your target sectors the day it’s released, and tracks the terms you care about. For example, if three retail CFOs in one quarter start crediting “AI agents” for flat headcount, budgets for automating store operations are forming. You get the quotes, the trend and a list of companies to call.
20. Feature flags in shipped code. Every night, the agent downloads competitors’ public JavaScript and compares it with the previous version. New flag names like enable_agent_mode_beta reveal unreleased features weeks before launch.
21. Shipping records. Every month, the agent pulls public bill-of-lading data on competitors’ imports. It shows their suppliers, their import volumes, and sudden jumps or drops that can signal a launch or trouble.
22. Help-center churn. Every week, the agent captures competitors’ help centers and logs new and rewritten articles. A burst of troubleshooting docs on one feature gets flagged as their pain point and added to your sales battlecard.
23. Domain verification records. Every week, the agent checks the public DNS records of competitors and target accounts. A new entry like openai-domain-verification or anthropic-domain-verification shows which vendor they just signed up with.
24. Layoff notices. The agent checks state layoff-notice filings (WARN notices) daily for competitors and target accounts. These often come 60 days before layoffs, so you see cuts before the press does.
25. Logo-wall diffs. Every week, the agent captures competitors’ customer logo pages and case studies. When one disappears, it flags the customer that probably churned so sales can call them.
26. Research papers as roadmaps. The agent tracks new papers by competitors’ researchers and summarizes what each implies about future products. This year’s papers often preview next year’s features.
Product management
27. Request evidence log. Each night, the agent reads every new support ticket, sales call and churn note, and tags each feature request. It logs the customer, a quote and the account’s revenue, so every request has a dollar total and evidence behind it.
28. Nightly persona testing. Each night, the agent runs 10 simulated users through your product, from a first-time admin to a power user on a slow connection. When one gets stuck, it files a bug with reproduction steps and a screen recording.
29. Closing the loop. When a feature ships, the agent finds every customer who asked for it. It drafts a personal note to each from their account owner: “You asked for this in March. It’s live.”
30. Zero-result searches. Every week, the agent groups every in-app and help-center search that returned nothing. Each group is a missing feature or a naming mismatch, ranked by frequency, with a suggested fix.
31. What customers build on your API. The agent scans public GitHub repos that use your SDK and summarizes what people are building. You find use cases you never designed for.
32. What sales promised. The agent scans every sales-call transcript for roadmap promises like “that’s coming next quarter.” It logs who promised what to which account, so product knows the commitments made in its name.
33. Churned users’ final sessions. When a user churns, the agent reviews their last three sessions and writes up what they were trying to do and where they got stuck. Each month, it summarizes the common patterns.
34. What retained users do in week one. Every month, the agent compares what retained users did in their first week with what churned users did. It reports the actions that best predict retention, so your “aha moment” stays current.
35. Error messages that create tickets. The agent matches the error messages in support tickets to your product’s error text and ranks them by ticket volume. It drafts clearer rewrites for the worst offenders.
36. Competitor 1-star reviews. Every week, the agent pulls competitors’ 1- and 2-star reviews from app stores and G2, and groups them by complaint. Their unsolved pain is your opening.
37. What reps demo and skip. The agent reviews demo recordings, logs which features each rep shows and which they skip, and compares that with win rates. The features reps skip don’t sell, whatever the roadmap says.
Design
38. Design system checks. Each night, the agent crawls production and compares every component it finds against your design system. It flags off-system buttons, colors and spacing, with screenshots.
39. Competitor onboarding archive. Every week, the agent signs up for competitors’ products with a fresh account and screenshots every onboarding step. You get a dated history of what changed.
40. Per-segment interfaces. The agent generates and maintains versions of key screens for each customer segment, like different dashboards and templates for banks vs. retailers. This used to be too expensive to build and keep up.
41. Figma vs. production. Each night, the agent screenshots shipped screens and compares them with their Figma specs. It flags drift in spacing, color and copy with side-by-side images.
42. Customer screenshots from support tickets. The agent collects every screenshot customers attach to tickets, including what they circled, and tags each by screen and issue. It’s your UI as it actually looks with their data.
43. Fifty-device sweep. Each night, the agent runs key flows on 50 real device and browser combinations and flags any screen that broke, with screenshots.
44. Worst-case data. The agent pulls real extremes from your customer data, like the longest company name, zero items or 10,000 items. It fills your designs with them to find what breaks.
45. Copy drift. The agent crawls every piece of text in your product and flags inconsistencies like “Sign in” vs. “Log in.” It shows where each version appears and suggests the standard.
46. Old-logo hunt. The agent reverse-image-searches the web for your outdated logo and brand colors on partner sites, directories and event pages. It emails each site owner the new assets.
Sales
47. Reviving lost deals. Every month, the agent rechecks each lost deal’s reason against what has changed: a new budget cycle, a new executive, new funding, or a feature that has since shipped. It flags the deals worth reopening, with a drafted note.
48. Customers’ job postings. The agent watches customers’ job postings. A post for doing manually what your product automates is an upsell, and a post for building it in-house is churn risk, so you know six months early.
49. Live prototyping on the call. The agent listens to the call and builds a mockup or working prototype of what the buyer is describing. The seller shows it before the call ends: “Is this what you meant?”
50. Answers routed in real time. When the seller can’t answer a question, the agent checks the knowledge base, then messages the right expert in Slack if it needs to. The answer reaches the seller before the call ends.
51. Live reference matching. When the buyer asks “Who else like us uses you?”, the agent finds the closest customer reference by industry, size and use case. The seller offers the intro on the spot.
52. Quote ready before hang-up. The agent listens for scope details like seats, modules and timeline, and builds the quote from your price book as the call goes. The proposal is ready for approval by the end.
53. Forwarding detection. When your proposal gets opened by people the seller hasn’t met, the agent looks up who they probably are at the buyer’s company. It alerts the seller to the new stakeholder.
54. Prefilled procurement forms. The agent fills in the buyer’s vendor forms and security questionnaires from your past answers. It flags anything new for a human, so procurement doesn’t add weeks.
55. No-show rescue. When a prospect misses the meeting, the agent sends two new time slots and a 90-second personalized video within 10 minutes.
Coaching, leadership and management
56. Promise tracking. The agent pulls every commitment you make from meeting transcripts and tracks whether each one got done. It shows your follow-through rate over months.
57. Calendar vs. priorities. Every week, the agent tags each meeting on your calendar against your stated priorities. It shows how much time each priority actually got, and which meetings served none of them.
58. Hidden load-bearing people. The agent analyzes Slack to find who answers the most questions outside their own team. They’re your real infrastructure and your biggest flight risk.
Recruiting
59. Readiness signals. The agent watches target candidates for signs they’re open to moving: vesting dates, a departed manager, a reorg, layoffs, or new GitHub or LinkedIn activity. It alerts the recruiter with the reason.
60. Acquisition retention cliffs. The agent tracks recent acquisitions and the key people at each acquired company. Since people often leave once their 12-month retention bonus pays out, it queues outreach for month 11.
61. Startup death watch. The agent tracks startups in your space that shut down or haven’t raised in 24 months. It lists their strongest people before they hit the market.
62. Specific referral asks. The agent checks employees’ networks against your open roles and sends each employee a short list of specific names with a one-click intro. With the employee’s sign-off, it can send the outreach from their email instead.
63. Live scorecard coverage. During the interview, the agent tracks which scorecard areas have been covered. It quietly flags the gaps while there’s still time to ask.
64. Interviewer calibration. The agent compares each interviewer’s scores with how their hires performed 12 months later. You learn whose “strong yes” predicts success.
65. Leaked interview questions. Every month, the agent searches Glassdoor and forums for your interview questions. It flags any that leaked so you can rotate them.
66. Work samples from real problems. The agent turns real past problems from your issue tracker into take-home exercises, with confidential details removed. It writes the scoring rubric from how your team actually solved them.
67. Job descriptions from your top performer. The agent reviews what your best person in the role actually does all week, from their calendar, issues and Slack. It drafts the job description from that, not a template.
68. Interview companion. The agent grades every candidate from the interview transcript against your rubric and gives a hire/no-hire recommendation. Over time, it compares its calls with your recruiters’ decisions and with how hires actually perform.
Health and wellness
69. Medical billing. The agent matches every medical bill against your insurer’s explanation of benefits and flags overcharges, duplicates and denials. It drafts the dispute or appeal for each.
70. Cancellation sniping. The agent checks a specialist’s online booking every few minutes and books the earliest cancellation that fits your calendar.
71. New research for your doctor. The agent watches for new studies relevant to your lab results and conditions, and summarizes each in plain language. It adds them to a list for your next appointment.
72. Experiments on yourself. The agent runs randomized, blinded trials on you, like caffeine cutoff times against sleep, so you can’t bias the result. It pulls the outcomes from your wearable and tells you whether the effect is real.
73. Adaptive gym coach. The agent watches your workouts through your phone camera and wearable, and gives form feedback. It adjusts your exercises, weights and intensity based on your form, heart rate and recovery.
74. Picking an open enrollment plan from your real claims. The agent pulls last year’s claims, prescriptions and visits, and calculates what you’d have paid under each plan. It recommends the cheapest plan for how you actually use care.
75. Procedure price shopping. The agent pulls prices from hospital price-transparency files near you, adjusted for your insurance. The same MRI can cost $400 at one facility and $3,000 nearby.
76. Calling to confirm doctors are accepting patients. Insurer directories are often wrong, so a voice agent calls in-network doctors and therapists to confirm they’re taking new patients. It books the first real opening.
77. One timeline from every patient portal. The agent pulls records from every hospital and doctor portal you’ve used. It merges them into one timeline of visits, labs, imaging and medications.
78. Family health history interviews. The agent schedules short calls with parents, grandparents, aunts and uncles, and asks about conditions, ages at diagnosis and causes of death. It compiles one family health history file for your doctor, and fills gaps as more relatives join.
79. Burnout early warning. The agent combines your wearable recovery data, sleep, meeting load and late-night email activity. It alerts you when the trend points toward burnout, before you feel it.
Family
80. Recall alerts. You list the baby gear, car seat and products you own once. The agent checks government recall databases daily and alerts you with the fix if anything is recalled.
81. Aging-parent protection. With your parent’s permission, the agent watches their bank, card and email accounts for gift-card purchases, wire transfers, new payees, scam emails and bills nearing their due date. It texts you both before the money leaves.
82. Recorded family stories. Once a week, a voice agent calls a grandparent, asks one question about their life plus follow-ups, and records the answer. Over a year, it assembles a family oral history book with the audio.
83. Kid quotes journal. You text the funny things your kid says to one number. The agent logs each quote with the date and your kid’s age, then lays out and orders a printed book every year.
84. Self-updating sitter packet. The agent keeps one sitter sheet of routines, medications, allergies, doctor numbers and bedtime rules, updated as things change in your texts, emails and pediatrician notes. It sends the latest version to whoever is babysitting.
85. Date night autopilot. Every month, the agent finds a night that works on both calendars, books a sitter from your list, makes the reservation, and sends you both reminders.
86. Printed photos for grandparents. Every month, the agent picks the best new photos from your camera roll and writes a short note about the month. It orders prints and mails a set to each grandparent, who don’t use Instagram.
87. Milestone tribute collector. For a parent’s 70th birthday, the agent invites 30 friends and relatives to record a clip, with prompts, and chases anyone who hasn’t replied. It trims every clip and edits them into one video ordered by era, with titles and music, a week before the party.
Personal experiences and consumer
88. Reservation and points sniping. The agent watches for hard-to-get restaurant tables, award seats and ticket drops. It books the moment one opens that fits your calendar and budget.
89. Trip rebooking. When your flight is cancelled, the agent rebooks you on the best alternative and updates your hotel and car, all before you’ve landed.
90. Price protection. The agent reads your order confirmations and watches the price of everything you bought. If a price drops inside the retailer’s or credit card’s protection window, it files the claim.
91. Waiting on hold. A voice agent calls the airline, insurer or utility, works through the phone menu, and waits on hold. It calls you only when a human picks up.
92. Registration at the exact minute. For summer camp, daycare waitlists, Global Entry and popular races, the agent fills in your details ahead of time. It registers the second sign-ups open.
93. Hotel rebooking. After you book a refundable rate, the agent checks the same hotel, room and dates twice a day. If the price drops, it books the new rate, cancels the old one before the deadline, and tells you what you saved.
Networking and relationships
94. Pre-meeting refresher. Before a meeting, the agent searches your past emails, texts and notes with that person. It sends you a brief: their kid’s name, the deal they were stressed about, what you promised to send.
95. Promise ledger. The agent scans your email and texts for every “I’ll intro you to X” or “I’ll send that over.” It nudges you before each promise goes stale.
96. Friend-group trip planner. The agent polls the group on dates, budget and must-haves, finds the weekend that works for everyone, and shortlists rentals. It runs the vote and books the winner. Then it tracks who paid for what and sends everyone a payment request to settle up.
97. Relationship strength scores. The agent builds a personal CRM automatically from your email, texts, calendar and LinkedIn. Each person gets a strength score that rises and falls with real interaction, and you get flags on important relationships that are cooling.
Learning
98. Gap detection. The agent notices concepts you keep looking up, and terms in your meetings you’ve never looked into. It queues a 10-minute primer on each.
99. Lessons from your own work. The agent reviews your meeting transcripts and turns them into a curriculum, like “what you learned from the last 20 customer calls,” with the key patterns and quotes.
100. Steelman feed. Every week, the agent finds the strongest argument against each position you posted publicly, with sources.
Home and real estate
101. Off-market alerts. The agent watches public records for probate filings, expired listings and owners who’ve held 20+ years in your target neighborhoods. It drafts outreach to each owner.
102. Maintenance from model numbers. You give the agent your appliance and car model numbers once. It builds a maintenance schedule from the manuals, checks for recalls, and orders filters and parts when they’re due.
103. Apartment hunting on a schedule. The agent checks Facebook Marketplace every two hours, and sweeps Craigslist, SpareRoom and other sites at 8am and 8pm. It screens out scams and emails the landlord the minute a good listing appears.
What "not possible" examples am I missing??
James Currier teaches startups at Stanford (average tuition == $67k/yr)
I went to James' house and asked him to teach us his class on the podcast, for free
Here are some of my notes:
1/ The ONLY Metric That Matters... SPEED
James’ fund (NFX) has invested $1.5B into startups. And of all the things they look for, the best proxy for success is SPEED. How fast the founder makes decisions, learns from those decisions, and iterates.
2/ You’re Moving Too Slow…And It’s Not Your Fault
We are trained in schools. Does anyone think school is particularly fast?
Then, if you’re smart, you get a job at a big successful company. Do they move fast?
If you’re trained in a slow environment, what you think is fast is actually slow.
Which brings us to big idea 3:
3/ You’re Must RESET Your SPEED BAR
James has a friend Mike Cassidy who did this for him. Mike understood speed. He sold one of his companies for $500M in just 500 days (!!)
When Mike would fundraise, he was told it takes 3 months to raise a round of funding. He would do it in 3 days. He stacked every meeting into a 24 hour period. He would tell them he’s meeting everyone, and whoever makes the best offer today gets it. Then when he got an offer, he’d tell everyone. This got the VCs to move fast, and make multiple offers. Within 3 days, he’d sign one and move on.
When hiring, he would bring a job offer to every interview. If he liked the person, he’d slide the paper over. They'd sign, and start the job the next day.
Every thing he did, he shaved days, weeks, months off the normal cycle time.
This broke James speed bar. He now had to move at Mike Cassidy speed.
His other friend created the viral Hot or Not app and sold it for $20M. The name of his company? 8 Days, Inc. Because he only would give himself 8 days to build and launch the product. SPEED.
The fastest way to move faster is to be around people who move way faster than you. It’ll pull your speed up.
4/ What Slows Us Down?
Our ego. Fear. Fear of being wrong. Fear of rejection. Being too precious about our idea. Thinking we’re special. You can’t move fast until you get over yourself.
When he advises founders, he will ask: “are you broken yet? Has the market broken you? You need to be broken first. Break your ego. Then once you’ve gotten over yourself, you’re ready to actually start moving.”
watch the full episode: https://t.co/Ie0sGjMQs4
back in London for a month, so here’s my contribution to the city.
introducing the London Food Money Matrix: a map of ~600 London restaurants, cafés and pubs and who owns them, whether that’s private equity, big corporates, or independent. look up a place before you go, and if it’s PE-owned it’ll show you an indie spot nearby.
why: went to Harry’s yesterday and the staff broke into Bella Ciao mid-dinner. since when is Harry’s a tourist trap?
so me and my friends looked it up - it was sold this year in a £1.4bn deal to an Abu Dhabi investment group.
https://t.co/abhZXNDaRd
This is a theme I consistently press on with founders, but @Alfred_Lin has done a service to all by making the point of speed so concrete.
Speed of execution when coupled with a bold vision substantially increases the chances of winning in a competitive market. Full stop.
@gamechangersvc #vc #founders #speed
Problema clássico de monitoramento: a inspeção é a cada 6 meses e o satélite passa a cada 5 dias. E no meio disso?
Na Tesar (https://t.co/HrSFDdQ2Hj) o twin sempre tem um valor pra qualquer variável, em qualquer lugar e instante, com o grau de confiança junto.
Por baixo é um filtro de Kalman (entre outros) em tempo contínuo por [lugar, variável]. O intervalo irregular entre medições já entra na conta. Quanto mais tempo sem medir, maior a incerteza:
"trinca: 2,3 mm ± 0,4, estimado, última medição há 42 dias, provavelmente estável"
"Our application is never blind due to the lack of data".
I mapped 67 companies doing AI transformation.
Labs, Big-3, AI-consultants, Palantir alumni, YC startups, company brains and agent platforms.
7 things I learned:
Why AI adoption is largely a psychological & cultural challenge, rather than a technology challenge.
I’ve just read the new book by @Gleb_Tsipursky, “The Psychology of AI Adoption at Work” as it’s one of the first books that offers in-depth advice about AI-enabled change through a human, rather than technology lens (& because I read his article on middle managers & AI in @HarvardBiz). The book describes itself as “a step-by-step tested guide for overcoming the biggest bottlenecks in AI adoption: human psychology & organisational culture.”
Some of the book’s key messages:
1) People talk about levels of “AI maturity” in organisations but there is no maturity curve. AI keeps evolving so there’s no end point. Adoption is a capability that needs constant renewal
2) Don’t underestimate the emotional & cultural side of AI-enabled change. AI touches identity; It automates writing, analysis & creative work that people see as part of who they are. Loss aversion, status quo bias & an empathy gap in leaders all shape the response
3) People support what they create. When frontline teams get involved in building their own AI tools, ownership lowers resistance, speeds up feedback loops & keeps capability inside the organisation
4) Engagement should be tailored. People hold different stances on AI & need varying responses: clarity on jobs for those fearful, data for the pragmatic, easy starts for the indifferent
5) Middle managers are key. They connect strategy with daily work & see the opportunities for change others miss. Hands-on experience of tests of change, peer mentoring & “premortems” help them lead with confidence.
6) Embrace sceptics – they make the work better. We should put sceptics on governance committees & get them to test AI tools for weaknesses before they go live.
7) Measure across several dimensions & frame measurement as learning. Time saved on its own is a misleading metric. One study found 36% of managers wasted more than half the time AI freed up for them
My main critique is that the book calls itself a playbook, but it offers approaches, not plays. The tools you would need to put them into practice (diagnostics, templates & worked methods on how to run each step) aren’t in the book. The introduction asks readers to build an adoption plan, guardrail map & KPI dashboard as they go, but it doesn’t provide templates or worked examples. In addition, the book would be strengthened if it pulled from established change & improvement methods, e.g., engagement frameworks, Plan-Do-Study-Act cycles & approaches to spread & scale.
I appreciate the content of this book. Its core insights are strong & important, from an author with a wealth of AI adoption experience. AI adoption IS change work. The principles that make any large-scale change initiative succeed also apply here: ownership, having a voice, psychological safety & learning as we go. Our challenge is to turn these principles into everyday practice.
Useful links for this post
The book:
Tsipursky G. The Psychology of AI Adoption at Work: From Resistance to Results. Georgetown University Press, 2026: https://t.co/dfPKtHhKXY
Companion resources referred to in the book: https://t.co/jqLs5DFeDH
Gleb Tsipursky interviewed about the book by Toby Corballis for the Wicked Problems podcast: https://t.co/5IRCq1PE3l
The original article I read from Harvard Business Review “Middle Managers Will Make or Break AI Adoption”: https://t.co/efEVtf7qla
The software factory is the next extinct wrapper: within 3–6 months, better coding agents will absorb much of its core orchestration value. The future is in a slimmer skills based approach with guidelines, guardrails etc. https://t.co/cqfMWaUjWK
If you're only working on vulnerability management, then you're only focusing on 1/3 of the problem.
If you prioritize breaking the attack paths around lateral movement and abuse of authorized access, then you're addressing all of the major attack vectors at once (vulnerabilities, phishing, stolen credentials, social engineering, software supply-chain, etc).
People love to focus on finding and fixing vulnerabilities because it feels good in the same way that running laps around a circular track feels good. You're moving fast and having predictable, repeatable tasks, but not actually getting anywhere.
We are witnessing the early years of AI system architecture. Every layer of the stack is evolving rapidly and at the top sits agent harness. Much of the engineering effort have gone into agent frameworks, runtime and infrastructure, leaving the design of the harness itself (e.g., tool design and context management) mostly open. This post is my attempt to summarize a few design principles of harness.
Thousands of Google engineers use this open-source plugin to draft specs with deep-context agents and stress-test them with zero-context subagents before writing code.
https://t.co/opZKWgxYmF
If you want to really understand AI agents, spend some time on this new site.
You won't regret it.
https://t.co/P6JBp2CFmN turns the whole subject into a map you can explore, and it's honestly brilliant.
Every page, term and tool is a bubble.
The lines show what depends on what, so you can actually see how the pieces fit together.
here's what's inside:
> a 7-module course, from a single prompt to a production agent
> 5 handbooks, each exploring one layer and ending with a build you can finish in an evening
> a 65-page hands-on project where an agent builds and maintains a second brain for you
> a glossary of 57 terms and 100+ tools
130 pages. 1,236 connections. 9-10 links per page.
It's FREE, written in plain Markdown, and the repo (in 🧵↓) is public too.
Apple has sold tens of millions of AirTags.
My buddy Dan bought four of them two years ago. He clipped one to his keys, one to his luggage, and then tossed the other two in a kitchen drawer and forgot about them.
Last month, he lost his wallet at the airport. He opened the Find My app and just stared at the screen, because he had no idea how any of it actually worked.
A guy standing next to him, a former Apple Store Genius, glanced over and said: "Bro, you're using 15% of that thing."
Over the next ten minutes, he showed Dan every setting he had never touched. Dan found his wallet in 40 seconds.
Here are 10 AirTag features and settings most owners never find, in the exact order he showed them:
What a time to be alive.
I’ve been playing with AI agents of late and I’m convinced beyond doubt that in just under few months from now, most SMB & mid-tier companies will hollow out their finance teams and run the entire function with just one Finance Engineer or max two. This person will be someone with finance domain knowledge + ability to prompt and vibe code.
Eventually this will happen across all functions and lead to one of Peter F Drucker principles of management becoming true 👇
“Because the purpose of business is to create a customer, the business enterprise has two—and only two—basic functions: marketing and innovation. Marketing and innovation produce results; all the rest are costs."