I attended many discussions, conferences and panels in the past few weeks.
Oftentimes executives are reluctant to apply AI agents to a broad set of systems. They are typically keen on removing write access and use AI to search and extract expert analysis.
Another key takeaway is that enterprises are reluctant to move away from their familiar interfaces. Companies that use Slack, Snowflake, Salesforce or similar UI portals will be reluctant to switch. Applying your AI solution as an app or extension will be the right way to spread it across a wide customer range.
The security environment has not changed and GDPR, SOC2, or HIPAA still apply to many environments. The successful consulting businesses go further in this space. Many companies that already use a cloud environment will stick to it. They require consultants to include any dashboards, reporting, and search stacks to run within their cloud environment.
SAAS is not dead and won't scale back. The wide web of communicating and cooperating startups will remain. However, their solutions will continue to spread fast as essential addons that rely on superior support, low cost, and quick ramp-ups for small businesses.
The main AI platforms were able to gain ground and keep their recurring customers. Claude Code from Anthropic, Copilot, OpenAI, Grok, Gemini are household names. The IPO of SpaceX and Cerebras will help to assess the real ROI on AI investment. The consensus in the industry seems that besides quick wins the models and companies that have the longer tenures will win the race for the market.
The early advances of some solutions like Cursor and Windsurf are fading. It seems that solutions that cooperate well integrating with many models will likely thrive over the ones that become silos. Many developers favor GLM or Grok over SWE or Gemini these days due to their strength and lower costs.
The interesting advancements to watch are change management. GitHub is traditional, Huggingface focused on researchers originally. An architecture and interface that allows versioning of models, SQL data, raw documents, and prompts are still on the TODO list of the industry. There are promising experiments like the versioning from TiDB. Our jetstream example describes the requirements from modern change management tools.
It is worth looking around if you use Cortex AI, Streamlit, Copilot, or Document AI. There are tremendous opportunities in finding established companies with an AI-native stack such as tray.ai, Cloudflare, or Replit. Tenure comes with more data that provides an edge in quality and fine grained design capabilities.