On resilience

Of the many factors that determine the success of a startup, resilience is a mysterious x-factor: it is both one of the hardest to predict and yet one of the most impactful on the ultimate outcome.

Resilience alone does not guarantee success. There are challenges and errors and bad luck that are great enough to overcome even the most resilient team.

I think, however, that the absence of resilience is enough to guarantee failure. Every startup that survives long enough to achieve success goes through at least one near death experience. At least one major customer disaster. At least one round that nearly fails. At least one human capital crisis. Health problems. Team crackups. Realizations that teams that fit together no longer fit. Without resilience, any of these inevitable crises would be enough to sink a company.

Resilience is optimism. To found a company is to bet on an unlikely outcome with an optimistic outlook. Of course the odds are low. They are tiny after the pre-seed round. They are not much better after the seed or the Series A. Resilience without optimism is impossible. The cold statistics will crush anyone who can not muster optimism.

Resilience is naivete. In the immortal words of Mike Tyson, “everyone has a plan until they are punched in the face.” But in startup land, founders are punched in the face again and again….and again. It keeps happening. I don’t know a founder - even of the most successful startup - who doesn’t feel regularly as if they have been in a knockout match with Mike Tyson himself nearly every day. The plans only go so far - which is not very far at all. The large outcomes are rarely the result of a linear progression down a pathway defined in a some plan. They are almost always a bit of a random walk - sometimes with closed eyes - from one moment to the next between knock-out blows and worry.

Resilience appears uncorrelated with commercial traction, fundraising, or valuation. I know founders with close to nothing who refuse to give up. I do not question them, nor do I accuse them of stubbornness. They act from conviction and self-knowledge. It is beautiful and honorable, even if it will likely end in failure. Most startups end in failure. I also know founders with revenue and access to capital who are default alive who can not muster the will to go on. I do not blame them, nor do I accuse them of capitulation. They are merely human and exhausted and need to rest. Sometimes it is wise to end the journey, sometimes it is wise to continue. Resilience, however, is orthogonal to wisdom. It is neither wise nor foolish. Resilience has a logic of its own.

As an investor, I bring some capital and some encouragement and perhaps some advice, but I can not bring resilience. That can come only from deep within. Resilience is the ability to maintain conviction as challenges are successively overcome. Some founders overcome challenge after challenge, but - in so doing - accumulate a manifest of fears that they eventually find overwhelming. Resilient founders seem to take every challenge overcome as a deepening signal that future challenges will also, in turn, be overcome.

We look for resilience constantly, but we find it in unexpected places. We look for signs of it in a founder’s history, in the way they tell their story, in their tone of voice, in the ferocity behind their eyes, in the quiet confidence of their hard-earned conviction. It appears unexpected after tragedy strikes a team member or after acts of god disrupt carefully laid plans.

Resilience is character. It is deep. It can’t really be taught, but it can be practiced and learned if one is self-aware.

In the land of startups, resilience is destiny.

Gil Dibner

Could the future of software be fluid
How do we get the best of AI without losing the soul of software?

The future belongs to young missionary teams
Why it makes more sense betting on youth in the current moment

The AI-native enterprise playbook
Ten real-time observations on a rapidly evolving playing field

No more painting by numbers
It’s the end of the “SaaS playbook.

ENTERPRISE/TECH NEWS

The rise of routing. Large AI buyers are actively routing work to cheaper models, optimizing prompts, and mixing vendors as flagship model pricing rises. Yes, demand for top-end models is still strong, but the unit economics are forcing real behavior change. This reinforces that cost-aware orchestration, open-source compatibility, and workflow-level ROI matter more now than raw model prestige.

Noam is out. Noam Shazeer, one of the key researchers behind the transformer era, is moving from Google DeepMind to OpenAI. The event itself is straightforward, but it is another reminder that frontier advantage is still deeply tied to individual talent concentration. (Made even more clear by the drop in Google’s stock price on the news.)

Open models go long. Z.ai released GLM-5.2, an MIT-licensed model built to sustain coding and research agents across a one-million-token context, with new architectural and reinforcement-learning techniques designed for extended, tool-heavy tasks. Z.ai reports that it is the strongest open model on several coding benchmarks, scoring 81.0 on Terminal-Bench 2.1 versus 85.0 for Claude Opus 4.8. The significance of this is not to be missed - if you remove Fable (which is unavailable), GLM-5.2 (Max) is the #1 model in the world for frontend coding. This is a huge moment. OSS has caught up with proprietary, and China has caught up with the US, in this very important domain.

HOW TO STARTUP

Inference gets its unicorn. Baseten raised $1.5B at a $13B valuation after growing revenue 20-fold, betting that companies will increasingly run cheaper open models on independent infrastructure rather than buy everything from OpenAI or Anthropic. The round is a striking validation of the idea that inference (not model training) will become one of AI’s largest markets. This also points toward a more modular stack: models may commoditize, while the layer that optimizes, serves and scales them still captures enormous value.

Security for agents. In this Latent Space conversation, Gray Swan cofounders Zico Kolter and Matt Fredrikson argue that AI security is not simply cybersecurity with an LLM attached: agents ingest untrusted data, access private systems, and can be manipulated into unsafe tool calls, while larger models do not automatically become more robust. Their answer is a new stack of automated red teaming, runtime guardrails, agent-native permissions, and eventually insurance and compliance. This suggests agent security will become its own infrastructure category and that deployment may be gated as much by trust and controls as by model capability. 

PostHog for vectorDBs. Sacra’s claim is that Turbopuffer has become the “PostHog of vector databases,” reaching roughly $100 million annualized revenue while expanding from vector search into full-text and hybrid retrieval. This is a story of a developer-first infrastructure company turning cost and self-serve distribution into rapid adoption, and it’s a strong example of how agent-era infra winners may look more like pragmatic product-led businesses than pure research stories.

HOW TO VENTURE

The applied AI playbook. SpaceX’s agreed $60B acquisition of Cursor is arguably the first mega-scale outcome at the AI application layer, validating a playbook built around deep domain focus, model routing, selective proprietary model development and relentless distribution. Cursor tuned every part of its business to keep gaining ground even as frontier labs pushed directly into coding, offering the clearest at-scale template yet for how applied AI companies can win. The next major opportunity may be the enterprise “control plane”: governance, auditability and continuity across models that lets organizations adopt AI without surrendering control.

CruxOCM
Technical Product Manager (Vancouver, Remote)

FalkorDB
Core Developer (Tel Aviv, Hybrid)

Jurnee
Founding BDR (Paris)