The harness margin gap: 71% cost reduction with no accuracy loss

Woman working at desk with coffee

This week’s signal from the founder beat: the companies that win the next software cycle won’t just pick the best model. They’ll build the best harness around it.

The Harness Margin Opportunity

The right prompt and routing layer cuts the cost of the same AI result by 71% without any accuracy loss. That gap is the product opportunity. State-of-the-art models are expensive by default, but a well-built harness makes them viable for everyday product use. Building one requires three things: deep knowledge of the customer’s workflow, a curated set of evaluations, and a system for automating iterative improvement.

AI-Native Companies as the New Kingmakers

Revenue scale no longer carries the weight it used to when raising a round. According to one recent piece, what VCs actually want to see now is which AI-native companies use, love, and publicly commit to your product. These customers are the hardest to win, which makes them the strongest leading indicator of real product-market fit. Revenue can be manufactured through discounts and enterprise deals. Loyalty from a company that builds AI-first cannot.

3D rendered ai text on dark digital background

Pulley Shuts Down, Sends Customers to Carta

Pulley raised more than $50M to take on Carta in the cap-table management space. It’s now closing on December 8, 2026, and directing its own customers to Carta. A former employee suggested the real competitive threat wasn’t Carta at all: AI-improved spreadsheets were quietly eating the category from below. If you’re a Pulley customer, you have roughly twelve weeks to migrate. Check now which records transfer automatically and what needs a manual export.

HubSpot’s 12 Months of AI Search Data

HubSpot ran a year of experiments on getting found in AI-powered search. The findings are concrete. Adding an llms.txt file had no measurable effect. Meanwhile, 92% of HubSpot’s highly specific industry and use-case pages eventually earned citations in AI responses. A page-speed overhaul drove a 1,600% increase in AI crawler visits. Qualified leads from AI search grew 1,850% over the year. The important caveat: more crawler visits alone don’t confirm the strategy is working. Track whether those leads convert.

More From This Week

  • OpenAI Sponsored Agents: OpenAI is testing a format where users can talk with a business-sponsored agent before clicking through to a site. It’s also adding ChatGPT Ads integrations to HubSpot and Shopify. If you get access to the test, your product information and agent answers will need as much attention as your landing page.
  • Mercury Books: Mercury added AI-powered accounting that generates P&L, cash flow, and balance sheet reports from Mercury data and external financial platforms. Co-designed with accountants. Existing Mercury customers can get started in minutes.
  • Column’s four new products: USDC and USDT are now interoperable with the US dollar across Column’s payment rails. Column built its own card issuer processor, so customers can launch card programs with one integration. Customers can also issue USD accounts and cards to verified users globally and hold foreign currencies in individually numbered accounts connected to international instant payment systems including SEPA Instant.
  • AI app defensibility: If a user can get the same result from a blank ChatGPT window, the question isn’t your answer quality. It’s whether your app remembers their progress, organizes what comes next, or gives them something to practice against. Those are the moats worth building.
  • Fundraising timing: One practical guide on VC fundraising recommends scheduling pitch meetings close together so investors reach decisions at the same time. Start before you’re under six months of runway. A signed term sheet still leaves due diligence and paperwork before cash hits the account.
Stay on top of AI & Automation with BizStack Newsletter
BizStack  —  Entrepreneur’s Business Stack
Logo