6 min read · 11 sources

Your customer’s procurement team already knows what your next renewal increase is funding: their AI token bill. That’s the dynamic reshaping B2B SaaS right now, and it’s breaking the pricing model that built the industry.

$55.7M/year: the average enterprise now spends this on software, with every dollar of growth coming from price hikes, not new apps.

The Seats Model Is Finally Dead

Source: saastr.com ↗

For four consecutive years, SaaS inflation ran at 12–16.4% against a general inflation rate near 2.7%. The average enterprise now spends $55.7M/year on software—up 8%—even as the number of apps in use actually slipped slightly. Every dollar of that growth is pure price. 79% of IT leaders saw a price increase at renewal. 78% got surprise AI or usage charges baked in without asking.

The math is ugly: 45% of CIOs are funding AI from existing software budgets, and 54% are cutting vendor counts to make room. Only about 28 cents of each new AI dollar is fresh budget. Your next renewal increase, in other words, is probably what pays for your customer’s GPT-5 bill—and their procurement team has already done that arithmetic.

SaaStr’s three-model framework lands on usage-based, outcome-based, and agent-based as the replacements. The agent-based model is the newest and most interesting: pricing tied to what autonomous agents do on your platform rather than how many humans log in. For engineers in B2B SaaS, this isn’t academic. Pricing-packaging decisions now directly affect how AI agents evaluate, select, and pay for vendors—a new kind of procurement that runs 24/7.

What Universities Should Actually Teach

Source: paulgraham.com ↗

Paul Graham’s latest essay takes aim at the entrepreneurship curriculum boom. His argument: the hard part of a startup is product—knowing what to build and being able to build it. Everything else is execution detail.

YC’s own data backs this up. Harvard alumni apply to YC at roughly twice the rate of Yale and Princeton alumni—not because Harvard produces better founders, but because starting a startup feels normal there. The culture and exposure matter more than the coursework.

His two prescriptions for universities are modest: make starting a company feel normal, and give students more time to work on their own projects. The implication for hiring and founder evaluation is direct: demonstrated ability to ship beats pedigree every time.

When to Actually Listen to Investors

Source: speedrun.substack.com ↗

Max Mullen—Instacart co-founder, now early-stage investor—shared a framework with a16z’s Farered Mosavat for sorting investor advice into three buckets. The Speedrun writeup distills it cleanly.

Science: Decisions with a right answer. Conversion experiments, pricing tests, go/no-go choices on features. Great investors have seen this movie before and can actually add value here. Take the call.

Religion: Culture, values, mission. Many valid answers. No outsider—no matter how experienced—should be telling you what your company stands for. Ignore.

Art: Timing, taste, product instinct. Also ignore outside advice. Your gut on a product decision after years of shipping is more calibrated than a board member’s pattern-matching.

The newsletter also previews Speedrun’s Global Founder Sprint for SF Tech Week (October 5–11), with travel sponsorship, 1:1 investor sessions, and VIP access for non-US founders. If you’re outside the US and fundraising, it’s worth a look.

The Discovery Meeting Is Not a Demo

Source: docs.google.com ↗

A Google Doc circulating among founders puts a fine point on something most sales processes get wrong: discovery meetings are not demos.

The purpose of a discovery meeting is to determine whether future communication is worthwhile—whether your organization can actually solve the other organization’s problem. Founders show up and give demos. They should show up and ask questions. Establish whether you can solve it and how you plan to do it. Everything else is theater.

Finding Product-Market Fit Again (Sorry)

Source: newsletter.posthog.com ↗

PostHog’s Charles Cook has a brutal post for established companies: the PMF game never ends, and AI-native teams are coming for your territory with lower build costs and more aggressive iteration cycles.

His playbook: write a “death page” enumerating three to four concrete threats, fund a three-to-five-person attacker team that reports directly to a founder—not the org chart it disrupts—and score the bet on seed-stage terms rather than business-unit metrics. PostHog is running this playbook live with PostHog Desktop, now in open beta, built around multiplayer spaces and generative UI artifacts.

The structural insight: internal teams get crushed by incumbent overhead unless they’re explicitly shielded from it. Founder reporting line, seed-stage scoring, zero shared infrastructure where possible.

The Full-Stack Trap in Robotics

Source: unbeatenpath.substack.com ↗

A post from Visionaries Tomorrow recounts a near-miss founding error: the team bought US-made Rover Robotics chassis (the same one Coco used) instead of cheaper Segway Robotics units, betting that hardware would commoditize and software would be the moat. Segway subsequently moved up the value chain into full integrated mobile robots with proprietary software, and Coco became a Segway customer instead of a competitor.

The lesson generalizes into the standard Chinese full-stack playbook: enter at a low-IP layer, then capture the higher-value integration and software layers above it. For robotics engineers and founders, the argument is direct: don’t outsource the hardware. Own the full stack or get commoditized by someone who does.

Compounded Mastery as a Moat

Source: prc.beehiiv.com ↗

Park Rangers Capital updates a 2023 thesis with an argument that sounds obvious until you sit with it: in 2026, when AI has made building software cheap and capital is abundant, the founders who win are those with “compounded mastery”—domain expertise, trust, audience, network, and pattern recognition built over years before the fundraise.

The claim: this asset can’t be bought or AI-generated because it lives in other people’s memory. VC underwriting modeled on the 19-year-old dropout is increasingly mispriced. The implication for engineers considering founding roles: your credibility in a specific domain, earned before you raised a dollar, may be the actual defensibility.

OpenTag: Self-Hosted Agents for the Chat Surface You Already Use

Source: github.com ↗

CopilotKit dropped OpenTag, an open-source self-hosted knowledge-work agent for Slack and Microsoft Teams. The agent can turn inputs—CSVs, for instance—into native Slack charts, Linear issues gated by approval workflows, and cited research briefs.

The stack: Node.js 22+ with pnpm, Python 3.12 with LangGraph. Deployment options include Railway (two-service template in .railway/railway.ts) or AWS ECS Fargate with Secrets Manager and Datadog log forwarding. Container images publish to ghcr.io/copilotkit/opentag-agent and /opentag-runtime. It ships with a managed channel runner (free) or a self-hosted runner via the open Channels SDK, with Discord, Telegram, and WhatsApp support on the roadmap.

For engineers, this is a reference implementation for wiring AG-UI agents into chat surfaces with durable state, approval gates, and generative UI—all without sending your data to a third-party API.

Apple's Quiet Sign in with Apple Migration

Source: developer.apple.com ↗

Apple announced (August 24, 2026) that later this year, new Sign in with Apple relay addresses will be issued on private.icloud.com instead of privaterelay.appleid.com. Existing addresses continue to forward without interruption. iCloud+ Hide My Email addresses stay on icloud.com after community feedback pushed Apple to keep that domain.

The engineering action: update account systems, email validation logic, and allowlists to accept both privaterelay.appleid.com and the new private.icloud.com domains. If you have any rule anywhere that matches privaterelay.appleid.com and rejects everything else, it will start rejecting real customers when new addresses begin issuing.

Bootstrapped Is Back, Maybe

Source: news.crunchbase.com ↗

A Crunchbase piece makes the case that bootstrapped businesses are gaining relevance, driven by focus on solving existing customer problems rather than chasing risky capital. The reasoning tracks: when AI reduces build costs, the advantage of venture-subsidized burning recedes, and the discipline of profitable unit economics becomes a feature rather than a constraint.

The piece doesn’t break new ground, but the trendline—capital more expensive, build cheaper, customer scrutiny higher—suggests the bootstrap calculus is improving for a specific founder profile: someone with revenue, domain credibility, and low burn.

The thread connecting most of these stories: the tools to build are getting cheaper, which means the moat is shifting from access to capital toward everything capital can’t buy—trust, reputation, domain depth, and the product instinct that comes from shipping, not studying.