AI Made Code Cheap—But Most Software Moats Didn't Collapse
Before AI, scaled software already had cheap replication costs; their real moats were habit, network effects, and proprietary data, not code scarcity.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Company Re-acceleration Is a Low-Probability Event
In response to Marc Andreessen's claim that ‘companies are either bonds or options’, Luca agrees the direction holds. Apple with iPhone and Nvidia with the AI wave achieved rare re-accelerations, but such shifts are mostly invention rather than inevitable: if you replayed history ten times, roughly eight times probably wouldn't happen—either someone else gets there first or it simply never succeeds.
— Luca FerrariLong-Term Cash Flow Compounding Over Current Valuation
Luca admits Bending Spoons could have used different capital structures and management approaches multiple times to appeal to investors and command a higher valuation, but instead chose to keep optimizing free cash flow's long-term compounding rate. His logic: valuation multiples are just one factor in multiplication; as long as the cash-flow factor is ten times larger, even a lower multiple produces a bigger final result.
— Luca FerrariDisruptive Innovation and Incremental Improvement Demand Choosing Sides
Luca believes pursuing disruptive invention and pursuing incremental optimization require completely different mindsets and cultures; a company should commit to one camp rather than trying for both and succeeding at neither. Disruptive innovation, when it works, can transform a company overnight—iPhone, Google search—but the median outcome is near-total loss; incremental improvement is more reliable, and in Bending Spoons' roll-up model—turning 100 of EBITDA into 200, then leveraging three times to buy more companies—compounds to near-hundredfold returns over time.
— Luca FerrariA Single-Product CEO's Fear of Loss Is Rational
Asked whether single-product founders irrationally overestimate downside risk, Luca says no: when teams won't make big bets, it's usually not that they've miscalculated odds by 20%, but simply that they dislike the ‘median outcome’. A VC holding 50 companies wants each to go all-in on the risk of hitting a wall, because if even four or five get through that narrow gate, the portfolio returns a hundredfold; but for a founder with one lifetime and one career, the median outcome of all-in sprint could shift from ‘something remains’ to ‘nothing at all’.
— Luca FerrariLeverage Capped at Three Times, Maturities Extended Beyond Five Years
Bending Spoons uses debt aggressively but enforces a hard constraint: leverage strictly below 3x (currently around 2.4–2.5x, historically closer to 2x), and even under severe adverse scenarios retains the ability to repay all debt before maturity; they refinance most debt every one or two years to keep maturities at five years or above, and have hedged 100% of debt against interest-rate swings. Luca's warning: almost every company that later went bankrupt sounded optimistic about its debt strategy before things broke, so even when numbers look safe, stay paranoid.
— Luca FerrariAI Made Code Cheap—But That's Not SaaS's Fatal Weakness
Luca doesn't buy the ‘AI made software cheaper, so these companies are done’ argument. He argues that most scaled digital products—even Salesforce—already had low replication costs before AI arrived (five fresh graduates can build an 80% clone in three months; the market is already full of WeTransfer knockoffs). What keeps these companies alive is first-mover habit and switching costs, network effects (Eventbrite's moat is 90 million monthly active ticket-buyers, not the ticket-selling technology itself), and proprietary data (Komoot's years of accumulated user-tested routes, invisible on satellite maps). Even if AI one day perfectly clones Airtable's code with a keystroke, someone still needs to own the code and fix it when things break—and that person's cost still exceeds paying Airtable directly.
— Luca FerrariSlack Bots Inside the Company Are Now Fixing Their Own Bugs
Luca describes Alt, an internal tool: an agent living in Slack with the same GitHub repos, Drive folders, and permission access as employees. It can be asked to do data analysis, fetch its own data (escalating to credentials and security teams if denied), plot charts and explain findings—delivering in minutes what once took many people and many hours. He gives a concrete example: Evernote's general manager asked the agent to check a support tool for whether other users reported the same bug, locate the root cause in the codebase, draft a fix, and @tag the engineering owner for review—all automated. At the company level, code was 0% AI-written two years ago, now 94%; in data science, fetching, cleaning, and charting work is at least 90% AI-handled; customer support is nearly entirely AI plus outside vendors.
— Luca FerrariBureaucratic Delay Accidentally Rescued Bending Spoons
Asked how founders should judge ‘should I quit’ versus ‘should I hold on three more months’, Luca admits he has no good answer. His pre-Bending Spoons startup was an AI app that auto-wrote journal entries—built too early, it pivoted constantly over three years, raising small checks, with users who didn't use it or pay. The balance sheet had roughly 250,000 to 300,000 dollars left, much of it counting on a Danish government tax refund. The government first said the money shouldn't be refunded at all, then said to wait another 12 months—enough to kill the company. His conclusion: the decision between ‘continue’ and ‘exit’ was made by bureaucratic delay, not by their own judgment.
— Luca FerrariIn their own words · checked verbatim
And if we do that right, we take the extra earnings and lever them up and buy more bigger companies, rinse and repeat.
Luca Ferrari1:03
So that's how powerful radical innovation is when it works, but it's highly, it's not dependable.
Luca Ferrari10:22
And my assessment right now is that our risk of defaulting on our debt is close to zero.
Luca Ferrari20:39
And rule number one to win and be that huge success is not having failed in the meantime.
Luca Ferrari21:40
So mostly software or digital products do not succeed because it's very expensive to replicate them.
Luca Ferrari25:49
We were at 0% written by AI two years ago. We're now, last time I checked, at 94%.
Luca Ferrari37:07
And so you could argue that red tape made the decision for us.
Luca Ferrari40:10
Figures
| Cumulative acquisitions | approximately 50–60 companies | 15:34 |
| Leverage cap | below 3x, currently around 2.4–2.5x | 20:39 |
| In-house tools | over 50 | 32:00 |
| Eventbrite monthly active users | 90 million | 26:51 |
| Enterprise software spending | 8–10% of total corporate spending | 28:59 |
| AI-written code percentage | 94% now, 0% two years ago | 37:07 |
| AI-completed data science work | approximately 90% | 37:07 |
Glossary
- SaaSpocalypse
- The market's period of concern after AI's emergence that SaaS companies could be undercut by low-cost copies, causing valuations to collapse.
How to listen
Founders and investors focused on SaaS moats, AI's impact on software, leveraged acquisitions, and roll-up M&A.
The opening company background and acquisition checklist can be skipped.