Search the archive

→ kiro.dev/startups

Terms and Conditions apply

CRACKS IN THE AI THESIS PART 3

Ramp's October AI Index reports that just 1% of firms in its data account for 80% of spending on OpenAI and Anthropic. That makes revenue unusually dependent on a small group of heavy users. Their September spending per employee was 3.1% below July's peak, even as overall token volumes rose. The index proposes falling model prices as the main explanation. If your AI revenue depends on usage, track what share comes from your largest accounts and whether their spending is recovering before forecasting revenue from token growth.

PRICE WAR IN THE CHIP AISLE

Five frontier AI models score within five points of each other on an intelligence benchmark, but the cost of running their tasks varies by 3.8 times. Google's Argon is cheapest at $1.99 per task. The comparison includes billed reasoning tokens and tests every model at maximum reasoning effort, so a similar score can come with a very different bill. Before choosing a default model, run the same sample of your application's requests through the candidates at the settings you plan to use and compare the cost of acceptable answers.

CYBERSECURITY INVESTORS PUT $6.8B TO WORK IN THE THIRD QUARTER, MOSTLY IN LARGER ROUNDS

Cybersecurity startups attracted $6.8 billion in Q3 2026, marking a 50% increase from the previous quarter, driven predominantly by larger funding rounds. Notable investments included Keyfactor's $1 billion round and Cyera's $400 million round, with both companies focusing on AI for identity and data security. M&A activity saw significant deals, such as Visa's $2.4 billion acquisition of BioCatch, emphasizing the industry's response to AI-driven security challenges.

CAL AI PAID UFC CREATORS THOUSANDS OF DOLLARS TO POST, AND GOT NO DOWNLOAD SPIKE AT ALL

A UFC fighter needs to track calories. Someone watching the fight might have no interest in doing that. Cal AI's co-founder blames that mismatch for creator deals that cost thousands and produced no download spike. The team now reads comments for people asking for help the app provides. It caps payment per thousand views at measured revenue per thousand views, favoring flat fees to keep viral upside. One creator went from roughly 10,000 views to 7 million. Before paying for that possibility, check whether the comments contain potential customers.

PUT A NAME ON IT

When project updates pass through a manager, leadership can lose the details needed to unblock the work. Nordhealth puts one engineer's name on each project and has that person post a weekly update in Linear, including changed dates and a link to the work. Reading roughly 30 updates takes leadership 30 to 60 minutes, leaving four to six projects needing closer attention. The owner can book Wednesday office hours directly with leadership. Start with a blocked project and check whether its named owner can get a decision without sending the question through their manager.

RIPPLING'S OUTBOUND GTM STRATEGY IS WORTH $17B

Rippling was exhausting its account lists with automated emails before it built an outbound team in 2022. Replacing sequences with people working the same accounts reportedly lifted conversion from roughly 1% to between 3% and 7%. Reps get paid for meetings held with decision makers, and account executives don't prospect. Managers supervise just four or five reps each. Selling multiple products to each customer helps fund that staffing. Before copying the team size, calculate how much you can spend to win an account, including the additional products customers actually buy.

BACKENDS ON VERCEL: PYTHON, CONTAINERS, AND DAEMONS

Vercel now offers generally available Python backends, containerized functions, and a public beta of Daemons for continuously running workers that pull jobs from queues or databases.

AGENT OBSERVABILITY ON THE CLAUDE PLATFORM

Claude's public beta of OpenTelemetry traces lets teams send agent turns, model calls, tool use, latency, and token usage into their existing observability tools to investigate failures and costs.

NAVIGE

Reduce agent costs through model routing, caching, budget controls, and safeguards.

ABSTRACT WARFARE: HOW TSMC MADE ITS OWN MARKET

TSMC made it possible to start a chip company without building a factory or depending on a manufacturer that sold competing chips. It manufactured for other designers and committed not to compete with them. A factory's huge upfront cost became something a designer could pay for as its business grew. TSMC initially survived on work existing chipmakers didn't want, while its future customers were still forming. Which expensive step must every new entrant in your market own, and could you sell that step without competing with the companies buying it?

MIT CLASS PROJECT TURNS INTO AN FDA-CLEARED TREATMENT FOR TREMORS

Encora's tremor wristband had people asking to buy it after a student demonstration, but it was nine years from that demonstration to FDA clearance. The MIT team chose the idea by a 16-to-one vote. Its proposer was the only dissenter. After initially working on Parkinson's, the company shifted to essential tremor, citing more patients and less drug-development activity. Its clearance covers adults with essential tremor. If your prototype attracts buyers, which patient group can you actually seek clearance to serve, and how long can you fund the trials needed to reach them?

I DESIGNED A NOSE GUARD FOR MY DOG AND GAVE IT AWAY. NOW 500 DOGS WEAR ONE

People wanted SnoutCover's dog nose guard but couldn't easily print and assemble the free design themselves. After trying to help them find local printers, its maker opened a shop. The first order arrived 16 days after the website went up. A year later, it had 395 orders across 16 countries. Customer photos led to better straps and custom fits. The design remains free, while the shop sells finished covers. If you give away a design, check which requests are for changes to the file and which are asking you to make it.

TRY CLARIFY FREE

ANTHROPIC REPLACED GIT BECAUSE GIT ASSUMES A HUMAN IS TYPING

Anthropic's daily merges rose from about 2,500 in May to more than 30,000 in September, making source control its biggest development bottleneck and leading it to build Scion, a replacement for git in its monorepo.

THE FREE UPGRADE THAT RAISED YOUR BILL

In a test of 191 constructed reading-assistant requests, a 33% cut in Sonnet's token prices produced a 4% higher bill with thinking off, as a changed tokenizer and longer answers increased billed tokens.

ON THE DEATH OF AMAZON'S WRITING CULTURE

Require human-written memos for hard-to-reverse decisions and hold presenters responsible for every sentence.

250 CEOS CONFESSED THEIR PAY. HALF EARN LESS THAN THEIR STAFF.

In Hampton's survey of 250 founders and CEOs, 47% paid a functional leader more than their own base salary, while none of the 58 venture-backed respondents reported more than $300,000 in compensation beyond salary.

Terms and Conditions apply, linked HERE [5].

THE “ALIEN” ECONOMICS OF AI MONETIZATION

Two-thirds of the companies on the Forbes AI 50 list use some form of usage-based pricing. AI's variable costs and unpredictable demand are causing today's leading companies to operate in ways that would seem alien to traditional SaaS businesses. These companies are iterating much faster on pricing, prioritizing market share over ARR, and pursuing multiple go-to-market strategies at once.

WHY THERE WILL BE MORE PRO RATA CONFLICTS

Early-stage funds increasingly exercise pro rata rights across multiple rounds as they grow, driven by outliers offering significant upside potential. The rise of secondary markets and financial vehicles like Fund of Funds and SPVs supports smaller funds in monetizing pro rata shares. Faster round pacing within a fund's investment period further contributes to potential pro rata conflicts.

38 THINGS WE'VE LEARNED ABOUT BUILDING AI PRODUCTS

PostHog has built 10 different AI products over the last four years. These include PostHog AI, PostHog's install Wizard, Replay Vision, and Scouts. PostHog's AI products are used by nearly 100,000 people every month. This post details what the company has learned from building them.

ARR VS GAAP REVENUE | SPOTTING “ARR” LIES

Businesses need to stay honest about what ARR actually is. If ARR doesn't ultimately lead to revenue and cash, then it's a fake metric. Most companies need to redefine or evolve their definitions of ARR for the AI era. ARR should turn into GAAP revenue, and it shouldn't take a really long time for that to happen.

HOW TO GROW A MASSIVE BRAND (FULL 2026 MASTERCLASS)

A founder's account and a company's account don't need to publish the same things. This guide assigns the founder the beliefs, lessons, and experience of building, while the company explains products and launches. It also distinguishes posts that attract new people from posts that build trust or encourage a purchase. That gives teams a clearer way to judge results: a post meant to introduce the company should not be assessed only by immediate sales. Interested followers can then move into email or a community for a longer relationship.

SEE HOW IT WORKS

RUN RECURRING JOBS ON STRIPE WITHOUT A CRON SERVER

Stripe Workflows now supports recurring schedule triggers. Users choose when a workflow runs, and Stripe starts a run at each scheduled time. Scheduled workflows pair well with list actions and loops. The workflow lives in the Dashboard, so the team that owns the process can read it, change it, and see every run.

GEMINI 4 ARGON AVAILABLE ON GOOGLE CLOUD, BUILT TO HELP SOLVE YOUR BUSINESS' HARDEST PROBLEMS

Gemini 4 Argon is now available on Google Cloud. Argon was trained to work with both depth and precision. It can maintain context across long, multi-step tasks and produce precise, verifiable results. Argon launches with an introductory rate of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off through to November 30, after which the standard pricing of $4 per million input tokens and $20 per million output tokens will apply.

OK, I MIGHT HAVE BEEN WRONG ABOUT BUILDING WITH A LEAN TEAM IN THE AGE OF AI

Chatbase is now at 40 people, and it plans to double its number of employees in the next 6 months. Building is now much easier, and the opportunity cost of not building more is huge. Companies that want to build more, grow more, and achieve more still need to hire because humans are still needed in many parts of the building loop. While this might change, it doesn't make sense to act like it isn't true right now.

CAPITAL STACKS FOR ATOMS

Capital allocation has shifted from bits-based businesses to atoms-based businesses over the last few years. Investors have been directing capital to overcome the physical-world bottlenecks of inference. There's an emerging cohort of atoms-based companies in the fields of energy, manufacturing, space, robotics, and beyond. These hard asset businesses are increasingly built and valued on bundles of cash flows rather than revenue multiples. This requires an entirely different approach to capital formation across the company lifecycle.

FRICTION

AI enables us to do the things we previously procrastinated over.

WHY MORE FEATURES WON'T SAVE YOUR STARTUP

Adding more features isn't a guaranteed solution for struggling startups.

THE DEATH OF VENTURE CAPITAL'S SHORT KINGS

Small VC funds face pressure to grow so they can keep investing in their companies and retain influence in later rounds.

DON'T DIAGNOSE YOUR OPPONENTS

Deciding that a colleague opposes your plan because they fear change does not tell you whether their objection is correct.

WE HIRED OUR WAY INTO GOING SLOWER

A founder has to protect their focus, attention, and execution on purpose, even when it's uncomfortable.

INCENTIVES DRIVE BEHAVIOUR

Angel investors prioritize founder stories and conviction over data, investing personal capital without external pressure.

EXPLORE STRIPE TREASURY

SOFTWARE'S CENTAUR AGE MAY LAST DECADES

Human engineers paired with AI coding systems are more effective than either by themselves. This fact is likely to remain true for the next decade or two. Software engineers have plenty of time to consider their next move. There's no need to panic and jump ship early when there's still time to live out an entire normal-length career.

THE DEVELOPER HIRING REBOUND FAVORS SENIORS

CNBC reports that software-development job postings rose even as overall US postings declined. Senior jobs supplied most of one measured increase, while employment among developers aged 22 to 25 remained under pressure. Founders should distinguish demand for experienced AI-capable builders from a broad recovery across the engineering talent market.

PART 1/3 OWNING INTELLIGENCE SERIES: HOW COINBASE BUILT, EVALUATED, AND POST-TRAINED A FRAUD AGENT

Coinbase built an LLM risk agent onto Onramp's existing ML model and rules. Online A/B experiments showed a 30% reduction in fraudulent transactions and 22% less fraud value. Coinbase then post-trained Qwen3.5-9B on its proprietary Onramp fraud dataset. The resulting model outperformed Opus 4.5 on all fraud detection metrics while delivering 55% lower end-to-end online serving latency.

“YOU DIDN'T DO IT RIGHT”

When OKRs don't work, how many times are you supposed to try again? WP Engine gave them four attempts, including paid training and coaching, and kept hearing that it wasn't doing them properly. What eventually worked was much smaller. One team picked a metric it cared about, set a loose goal, and followed it for a quarter before adding anything else. The metric also had to measure something the team could actually control, so people weren't being judged on a number their own work couldn't fix.

YOUR CHURN PROBLEM STARTED IN THE COMP PLAN

Your reps are hitting their targets, but a year later the customers are gone. One founder in this piece lost 30% of the previous year's new customers and wanted to know what was wrong. The answer starts with what reps get paid to sell. In a separate Airtable example, a birthday-calendar user counted as much toward the new-logo target as a team running a critical workflow. Airtable began scoring prospects against its best customers, giving reps smaller account lists and paying more for customers with room to grow.

Hire your first AI worker now →