TLDR: SAN FRANCISCOāDatabricks co founder Arsalan Tavakoli Shiraji says enterprise AI deals stall because deployments feel risky, not because models fail. Buyers now demand safe scaling, affecting pilots and contracts.
Key Takeaways:
- Enterprise AI moved from demos and pilots to safe broad deployment after years of experimentation.
- Tavakoli Shiraji says deals die when enterprises lose confidence in implementation risk, governance, workflow disruption, and infrastructure strain.
- Startups that reduce uncertainty and integrate cleanly win durable revenue, while novelty focused teams stall after early momentum.
The market is done being impressed. Enterprises want receipts that the rollout will not scramble teams, systems, or compliance, and they will make that their gatekeeper.
The market is done being impressed. Enterprises want receipts that the rollout will not scramble teams, systems, or compliance, and they will make that their gatekeeper.
Q&A
If pilots are no longer the bottleneck, what should AI founders instrument first to prove deployment readiness?
Track operational signals like workflow disruption, failure recovery time, governance overhead, and infrastructure strain during staged rollouts, then tie them to business outcomes.
Why do enterprise buyers treat model performance as necessary but not sufficient now?
Because high accuracy in a controlled test does not predict stability, auditability, and day to day adoption in real systems where incidents and process friction show up.
What happens to procurement timelines when governance and compliance exposure become central to buying decisions?
Deals slow down but get more predictable when vendors deliver clear documentation, controls, and rollout plans that legal, security, and operations teams can sign off on.
How can startups reduce uncertainty without turning every project into a long enterprise consulting engagement?
Pre package adoption with reference architectures, clear operating procedures, and measurable rollout playbooks that shorten the internal work enterprises must do to feel safe.
Could this shift reshape which AI startups win over the next cycle, even if their models are not state of the art?
Yes, because buyers increasingly reward teams that integrate cleanly, govern easily, and earn trust over time, which can outperform headline benchmark wins.
Top in AI
Google AI Overview keeps stumbling on letter counting
Frontier LLMs fracture on fact-check verdicts 67% of times
Claude Opus 4.8 ramps agent reliability while pricing stays put
Gemini, Claude, and ChatGPT unravel in AI radio trials
No comments yet. Be the first to share your thoughts!