An AI IDE can make engineers ship faster. That does not mean the team is getting better. NVIDIA saw 3x more committed code across 30,000 developers without more bugs, but the companies closest to that result had rebuilt the workflow around GitHub, Linear, review, and testing. Another dataset is the warning: epics finished 66% faster while bugs per developer rose 54%. If your AI dashboard only shows pull requests, it is probably hiding half the story.
A software factory is a coding pipeline where agents take a spec and return working code with little or no human review. The pitch borrows from lights-out manufacturing: the line runs while nobody is on the floor. HumanLayer tried it in July 2025. Three months later, a production bug dropped the team into code nobody had been reading, and the cofounder spent two weeks rewriting the core patterns. Agents can run the line. Someone still has to understand the factory well enough to know when it is drifting.
Duolingo did not sell better software to schools. Coinbase did not sell better ledgers to banks. Both moved something the gatekeeper controlled directly to the customer. AI founders now face the same choice. An AI doctor that can prescribe changes the market more than software that helps a doctor type notes. Software that runs a factory changes more than a dashboard sold to its manager. If your customer still needs the same gatekeeper after buying the product, you may be making the old market faster rather than building a new one.
Most founders cut onboarding. Cal AI made users do more of it, then built a multimillion-dollar subscription business. Before the price appeared, users entered their goals, body type, habits, and timeline, then watched the app assemble a plan. By the time they saw the paywall, it felt less like buying another calorie tracker and more like walking away from a plan made for them. That only works when the pain is real: every question must sharpen the product or increase commitment.
What does a healthy B2B pipeline actually look like? One founder with real revenue and Fortune 500 customers discovered that most of his pipeline was fake. This is the scorecard he wished he had. For a $100K+ product, each seller should book three first meetings a week. 60% should show real buying behavior, half of those should close within 90 days, and 95% of customers should reach the first retention milestone within a month. You can disagree with the numbers. At least they are hard to explain away.
2026.07.27·thephysicsofstartups.substack.com·8 min read·TLDR Founders This post shares how the team at Clay uses AI in their day-to-day. It covers how to generate on-brand decks from a single prompt, spin up custom apps for each customer, and clear the Salesforce chores nobody wants to touch. The post walks through Clay's full AI sales stack, the exact tools used, and 10 concrete ways the team puts AI to work.
The GitHub Copilot app is an interface for agent-driven development built natively on GitHub. It allows developers to move from issue to merge all in one app. Developers can run multiple sessions across every area of work with parallel workflows fully in view. The app can be extended with MCP servers, plugins, and skills.
Claude Opus 5 is now available on all platforms, priced at $5 per million input tokens and $25 per million output tokens. A Fast mode that runs around 2.5 times the default speed is available at twice the base price. Anthropic has also introduced mid-conversation tool changes on the Claude Platform and automatic fallbacks on the API. Details about what has been improved in Claude Opus 5 are available in the article.
After reviewing 15 of her own agent sessions, NextView's Melody Koh found that the agent was not wasting her time on the work. It was wasting it on corrections. The same mistakes kept returning because she had never told the model what "good" meant for quality or positioning. Her fix was to make those judgment calls upfront, let the agent handle the checkable middle, and review once at the end. If you correct the same mistake twice, it belongs in the instructions.
A junior engineer does not become senior by doing senior work. They get there by doing the routine work first: reading the codebase, seeing exceptions, and learning what "good" looks like. Coding agents can remove that apprenticeship while making output look better in the short term. The useful question is not "which jobs can AI replace?" It is "which capabilities disappear with the task?" Before automating work, ask whether it was also building context, taste, or the person who would catch the weird case later.
Sometimes it is wiser to eliminate the defaults within yourself.
Service with a Software - private tools that never get sold, built to make one service impossible to compete with - is the strongest position left for independent builders.
This fund now looks for founders who are good at StarCraft because they already know how to allocate scarce resources, sequence parallel work, and intervene before one bad unit loses the map.
Winning teams are quick to admit what they don't know, and quicker to help each other fill those gaps.
Patience is a form of arbitrage.
What made someone admirable in one era can look quite different the next. Founders must continually reevaluate what earns legitimacy in order to survive these cycles. One could opt to not play in this game, but this would go against the industry's broad desire to change the world. It would be bad for tech to be more culturally and socially isolationist at the exact moment it gained the most power over culture and society.
With every AI startup seemingly hitting $100 million ARR in record time, this article asks how much of that revenue is actually theirs. A company can book $100 million and send $90 million straight to Anthropic, OpenAI, or another model provider. Cheaper models do not necessarily fix it. Customers will expect the savings or bring the inference they already pay for. For your own startup, ask what they would still pay for if the model were free. That is the part of the product you own.
Kimi K3 is the latest in a series of open-weight models. Open-weight models are estimated to be used in 80% of startups. They represent 25% to 50% of the volume across major platforms like OpenRouter and Vercel. The gap between frontier and open-weight performance has compressed to 4 to 6 months and appears to be shrinking.
VROOM started with a convincing idea. Event planners carry a list of venues, vendors, and prices in their heads, so why not turn it into software for a $90 billion market? Four months later, the team learned that prices went stale, coordination was not that hard, and customers were really paying for trust. They spoke to planners, but translated trust into "information asymmetry." For your own startup, ask whether the expert's advantage still works after that person leaves. If it lives in relationships, you may be building a service.
Everyone watches funding announcements for sales signals. By the time you email the company, 30 other vendors are already mentioning its Series A. A hiring post may be better. A founder advertising for an SDR has just published the pipeline problem, approved budget, and roughly three-month timeline in public. Search LinkedIn posts from the past week for "we're hiring" and the role your product could replace or support. Focus on founder-written posts from companies with fewer than 200 employees, then show how you can produce part of the desired result before the new hire even finishes ramping. You are not trying to talk them out of hiring. You are arriving when the problem is urgent and the budget already exists.
Notion as code allows users to define and deploy an entire workspace in TypeScript. Users can build workspaces with coding agents, version-control the setup in git, and reproduce the same setup anywhere required. They can review every change and keep a clear history of how the work evolves in git. The same script can be used across dev, staging, and production.
Rogo Intelligence transforms firms' expertise into reusable institutional memory. It sits between a firm's knowledge and the AI systems that use it, forming fragmented information into structured and governable institutional memory. Rogo Intelligence transforms a firm's intelligence into durable infrastructure that is governed, permissioned, reusable, and continuously improving. This will allow firms to create an enduring advantage as frontier models continue to improve.
A new Zerodha employee knew the company does not push customers to trade, but could not explain why. For a 16-year-old business whose reputation rests on that decision, onboarding had preserved the rule but lost the reason. The same culture shows up in smaller choices. No 3 AM emails, no weekend work, and no random founder ideas derailing engineering. It can take almost two years for an employee to contribute properly, so burning people out at year two is expensive. Try asking a recent hire what your company refuses to do and why. If they only know the slogan, onboarding fails.
Everyone is hiring forward-deployed engineers now, but the title is starting to mean almost anything. An actual deployment means flying into a customer's operation, meeting everyone who touches the problem, and shipping a first version in about a week. Deployment one is ugly. By deployment ten, templates and integrations should turn most of the work into configuration. If customer ten still needs the same people and hours as customer one, you built a consultancy with a software demo.
Steve Ross started by renting his father-in-law's unused funeral limousines at night. He added parking lots, cleaning companies, plumbing, and other fairly boring businesses. That collection eventually bought Warner Bros. These companies were attractive because ambitious people rarely bothered to look at them. Constellation Software found a similar opportunity in tiny software businesses serving marinas, funeral homes and libraries. There may now be more than 38,000 of these vertical software companies. Perhaps for your own startup, look one or two layers beneath the industry everyone is chasing. The better business may be selling to the supplier nobody at the conference thought was worth meeting.
People open your product because they want something to happen.
Investors are now asking whether companies can become extreme outliers rather than whether they are strong companies.
This post looks at how to use agents to run Gumroad support from a phone.
Adoption and investment in AI is growing, but sustained impact on performance is elusive.
Founder-market fit, the pairing of a specific founder to a specific market, is the most durable signal in venture.
Sponsored
Organizations shape technology and technology shapes organizations. Each wave of technology changes what it means to build and run great organizations, and AI is no different. The firm of the past was designed around the bottleneck of human execution. The firm of the future will be designed around the bottleneck of human taste and judgment. This article looks at factors organizations can adopt to be smaller, faster, and deliver outsized value to customers.
Everyone says agents will be the main users of software by 2030. Okay, so what should the product actually look like? PostHog's answer is pretty simple. Agents should do the work, while the interface helps humans approve decisions and check what changed. If a person is clicking through a screen only to perform an action, give the agent a way to do it. Keep the screens where someone needs to judge the result, resolve something ambiguous, or stop the agent before it does something dumb.
Building an autonomous software factory feels the same as speedrunning a game. AI made building cheap, and speeding up the software factory lets you take more shots at building things. This post looks at how Amplitude built its software factory, which helped the company 3x its PRs. The faster a software factory is, the faster you can put more ideas in front of customers to figure out what works. They allow you to run more experiments and see good ideas surface much faster.
Many leaders want a forward-deployed engineering (FDE) team, but few know how to build one. The people who excel in the role are scarce, expensive, and hard to interview for. This article looks at what an FDE is, what makes a great FDE, how to find them, and more. It includes an overview of how other companies are running the FDE model and covers common misconceptions about the FDE bench.
Profitable product ideas often come from problems founders have personally solved, giving them insight, proof, and credibility. Filter those problems by links to money, time, or status, identify the root cause, and validate whether buyers will pay for the transformation.
People love to say startups die by suicide, not competition. That is wrong, especially in B2B. A lower win rate means less revenue, then cuts, worse morale, and an even lower win rate. That is why he doesn't care about 5% product improvements. A customer usually needs a painfully obvious reason to switch, like half the price or 10x the speed. Look at the deals you lost and ask what would have made the choice obvious without another sales call.
Sponsored
Ads get ignored on social media but not in TLDR! Reach developers, PMs, marketers, founders and other tech leaders where they actually
Buzz is a workspace where humans and agents build together. It is self-hostable, and an operator can serve many communities behind many domains or subdomains. Buzz feels like a team workspace, but under the hood, it's an event log with taste. It is a relay where humans, agents, workflows, git events, and project memory cooperate.
OpenAI has launched a program that offers small businesses virtual training, in-person AI academies, workflow guides, and partner-built tools for using ChatGPT across operations such as marketing, accounting, and ecommerce.
In a study of 1,094 US ChatGPT categories, only 15.2% had a clear owner. Most companies are still fighting over categories where ChatGPT has not settled on a default brand. Once a brand did become the default answer, it stayed first in 90.4% of next-month comparisons. Maybe don't publish the twentieth generic AI article. Pick the five questions you want your company to be named for and build around those.
Marketplace funding has become sharply selective: Pre-Seed and Seed valuations hold, but Series A rounds, capital, and valuations have all fallen. Founders increasingly need Seed extensions, stronger revenue, better liquidity, and higher-margin products to bridge an 11x revenue jump.
Most VCs say they invest in people over ideas, but this rarely gets tested since they join when there's an existing idea and market.
A customer can admit the problem is real and still do nothing.
2026.07.22·averageentrepreneur.substack.com·2 min read·TLDR Founders Hardware startups must make seven critical hires, focusing on manufacturing, deployments, supply chain, finance, sales, policy, and marketing.
Effective introductions can enhance reputation and create valuable connections.
AI support bots are one of the clearest enterprise AI use cases. The other commercially-advanced use case for AI is coding agents. The industry is seeing more and more usage of open-source Chinese models. The rise of Chinese models indicates that businesses want something that isn't being offered by the American model companies.
A $10 million financing can now be called pre-seed. In 2014, pre-seed usually meant $1 million or less and enough money to find the first signs of product-market fit. Now the label is often chosen so the company can still call its next, even larger financing the seed round. If an investor says you are too early or too late, ask what they actually mean. Checking size, ownership, valuation, and the milestone expected before the next raise will tell you more than the round name. "Pre-seed" certainly does not tell you how much money the company already has.