Five Themes from Tapestry’s ‘Threads’ Day
Last month, we held our second Tapestry Threads day, bringing together many of our portfolio, investors and friends for a day of discussion about the present and future.
Across six panels throughout the day, one narrative was ubiquitous. Whether talking about building companies in robotics, cybersecurity, autonomous transport or fintech - one change was cited by every founder: Artificial Intelligence is rewriting the startup playbook.
Our portfolio of Repeat founders have a unique vantage point: they have built startups before. This experience is incredibly valuable, enabling them to move faster and make less mistakes as they build again. But the game is never the same, and that is more obvious now than ever.
Up on stage, Tapestry VC’s partners debated the new highs and lows of startup building in an AI-first world with these founders: how AI is reimagining hiring and the organization chart; how the lines are blurring between hardware and software; how it is levelling the playing field across the Atlantic; and much more.
Afterwards, we felt these insights were too good to keep to ourselves. We have put down our five major insights from the day below, while respecting the confidentiality of our companies and friends plans and metrics. We hope they are valuable to you.
With thanks to our portfolio founders for their valuable insights & contributions to the afternoon’s discussions:
Manna Air Delivery - Bobby Healy
Sunrise Robotics - Joe Perrott
Huckleberry (acq’d) - Bryan O’Connell
Pointy (acq’d) - Mark Cummins
#1: AI Isn’t a ‘Vertical’, it’s the New Foundational Infrastructure
First up, a refrain we heard from Tapestry VC’s own Patrick Murphy at the outset of the day was echoed throughout the later sessions: "saying you’re ‘investing in AI companies’, is like saying you invest in electricity- or software-based companies. It’s no longer a meaningful delineator. All companies must be AI companies.”
The strongest teams are deeply vertical, with domain-specific insight and expertise; only since AI capabilities reached their present maturity could the technology be fully foundational. The distinction now is not whether a company is AI-enabled, it is how AI is being leveraged by the company.
In fintech, AI-native portfolio company Seapoint is automating onboarding & KYC to be completed in hours, not weeks. In logistics, Manna Air Delivery uses AI for dynamic path planning and collision avoidance. Cybersecurity firms are emerging to combat what Tracebit’s Andy Smith warns are an “an army of AI hackers - and that’s kind of terrifying”.
This is not feature-building. The best founders are building defensibility not through proprietary models, but by applying their domain mastery through the lens of this new AI infrastructure.
#2. Tearing Up the Start-Up Org Chart
Foundational AI isn’t just accelerating what startups build - it’s reshaping who they need to build it.
Across our panels, we heard examples of companies compressing years of work into months. At Seapoint, a five-person team built an enterprise-grade banking product in under a year. Another company mentioned by our panelists “went from zero to $7M ARR in 18 months… with seven people”.
At Tracebit, co-founder Andy Smith explained: "We can use LLMs to distinguish whether something looks like a decoy or not, and then use that feedback loop to make our canaries undetectable in their deployed environment." In other words, iterative agentic AI is automating product development.
This new paradigm isn’t just about headcount reduction. It’s about building teams where every human is paired with AI counterparts, creating "an army of co-pilots". Concurrently, the bar for talent has shifted. Fewer engineers are needed, but you need “weapons-grade” talent. In larger enterprises, a number of our speakers spoke of entire teams once responsible for data processing being replaced with autonomous systems.
It all leans into one panelist's belief that “early-stage startups now have the same software leverage as $100M+ revenue companies a decade ago”.
#3. Where Value Accrues: A European Reframe
One tension that surfaced repeatedly: the geography of value creation.
The instinct was to look to foundational models and US-based giants. But today, value is accruing across the application layer, and despite the persistent transatlantic funding gap, that value is being geographically democratised. Capital efficiency and dispersed talent pools are leveling the playing field.
Overall our panelists were optimistic: “Being here, starting an AI company in Europe… is a much closer parallel to starting an AI company in San Francisco today, than a SaaS company was in 2015.” Others had worked with companies where 100% of a UK- or EU-based startup’s revenue was coming from US-based customers, to the tune of millions in ARR.
Where historically the regulatory environment has been thought to have stifled Europe’s competitiveness & opportunity; the EU AI Act looks set to spawn a generation of applications with compliance-built-in defensibility.
Bobby Healy, founder of Manna Air Delivery, reflected that EU regulatory clarity and alignment presented a massive market creation opportunity for his business, whereas “the US is actually 2 or 3 years behind Europe” in his area.
#4. Synthesising the new Dataset
The bottleneck in the development of Agentic capabilities is data.
Where Generative AI has a near-infinite source of text & media, agentic AI requires rich inputs: data that lets it engage with a variable environment and make autonomous decisions.
Joe Perrott, co-founder of Sunrise Robotics, said: “The key problem in robotics is around the scale of the data… You scale the input 10x and the output only gets better by ten or twenty percent. But it keeps going, so if you scale it 1 million times higher, the performance goes up by a much higher percentage. You scale it 10 million times, it keeps going.”
Until recently, engineers have painstakingly surveyed physical spaces and captured processes in order to train autonomous capabilities. Alternatively, they have paid millions to minimum wage human workers to create specific training data. Today, simulation and synthetic data are filling the gap.
As Waymo has done for autonomous driving, the team at Sunrise is training their robots - not programming them - using “behaviour trees in a simulation”. The compute power is immense, but the productivity gains are even more so.
#5. A New Playbook Demands New Capital Models
If AI-native companies are built differently, they need to be funded differently too. “The playbook of what you need to look like to raise a successful round has changed,” one speaker said. “And it’s still unclear what that is.”
Many echoed the sense that traditional venture models may be ill-suited for changing demands. Molten Venture’s Nicola McClafferty put it plainly "our funding models will need to fundamentally change... there will need to be innovation in our capital markets and the venture capital ecosystem to support the level and type of companies that we’re going to see over the next 10 years."
Whether it’s the need for compute, talent, or regulatory readiness - capital intensity has been upended by AI.
For software development, capital intensity is tracking to nil (see #2 above). Entire products could now be built and brought to market by bootstrapping. But investors' value-add extends beyond deep pockets - and quantifying that contribution will be a novel challenge for the cap tables of the future.
