AI is reshaping utility growth. Discover key tendencies from Cosmos DB Conf 2026 and the way groups are constructing scalable, AI-native functions with Azure Cosmos DB.
Yearly, Azure Cosmos DB Conf affords a window into how trendy functions are constructed—not in principle, however in manufacturing at international scale.
This 12 months, the important thing theme from Cosmos Conf was clear: AI isn’t just one other workload. It’s essentially reshaping how functions—and knowledge platforms—are constructed.
Within the opening keynote, VP of Azure Cosmos DB Kirill Gavrylyuk described three key shifts driving this transformation, and we noticed them play out throughout each buyer story on the occasion.
The three AI shifts reshaping utility structure with Azure Cosmos DB
AI is making versatile, semi-structured knowledge foundational
AI functions don’t function on inflexible schemas. They function on prompts, reminiscence, and context, all of that are inherently semi-structured and evolving over time.
This essentially adjustments how databases should behave.
Knowledge platforms are now not simply programs of file—they’re changing into programs of reasoning, the place flexibility is essential to how functions be taught, adapt, and generate outcomes.
AI is dramatically accelerating the tempo of growth
AI, and particularly coding brokers, are altering how software program is constructed.
Builders are:
- Iterating quicker
- Transport extra continuously
- Scaling from zero to huge utilization immediately
As Kirill highlighted, builders can now not be constrained by strict schemas. Flexibility isn’t only a comfort—it’s what permits groups to maneuver at AI velocity. Databases want to satisfy the demand with serverless type issue, immediate and limitless scalability, superior built-in caching, and supply agent-friendly interfaces.
Semantic search is changing into a first-class question operator
The third shift is simply as essential:
AI functions require:
- Vector search
- Full-text search
- Hybrid search
- Semantic rating
These are now not “add-ons.” They’re core to how trendy functions perform.
Throughout Cosmos DB Conf, we noticed a transparent sample: groups are constructing functions the place retrieval, reasoning, and real-time context are tightly built-in.
OpenAI: Flexibility at planet scale
These shifts are most seen in what organizations like OpenAI are constructing.
Talking at Cosmos Conf, Jon Lee of OpenAI addressed how they’re working at huge scale—processing trillions of transactions and petabytes of knowledge—reinforcing that what issues most isn’t just scale, however the potential to evolve rapidly.
As Jon shared, trendy programs should have the ability to:
- Scale immediately from zero to huge utilization.
- Help schema-less design for fast onboarding.
- Allow 1000’s of builders to iterate concurrently.
“Crucial factor… is having the ability to scale from zero to hundreds of thousands of QPS, having the ability to scale from zero bytes to petabytes,” defined Jon, including that velocity and adaptability go collectively.
Now we have 1000’s of builders which are actively constructing merchandise… it’s actually essential to make it straightforward to onboard to databases actually quick.
That is precisely the world Kirill described: AI programs demand versatile knowledge fashions that evolve as quick because the functions themselves.
This highlights how Azure Cosmos DB helps dynamically evolving, large-scale AI workloads.
Vercel: The rise of serverless, AI-native functions
If OpenAI exhibits what’s potential at scale, Vercel exhibits how the form of functions is altering.
As Guillermo Rauch, CEO of Vercel, defined, AI is dramatically increasing who can construct software program—from hundreds of thousands of builders to probably billions of creators, lots of whom are utilizing brokers to generate functions on demand. Kirill underscored this level in his keynote when he said that greater than half of Azure Cosmos DB prospects are already utilizing coding brokers of their growth workflows.
In line with Guillermo, that is driving a structural shift towards:
- Serverless architectures
- Ephemeral functions
- Prompt scaling from zero to viral
Knowledge platforms should sustain. To assist this tempo, platforms want to supply:
- Constructed-in finest practices (knowledge modeling, partitioning, and optimization).
- Clever steering (agent abilities and automation).
- Actual-time suggestions on efficiency and value.
Talking on why he turned to Azure Cosmos DB, Guillermo stated, “I needed a system that gave me a cost-effective considering the place the developer writes a question they usually perceive its price.”
Builders want quick suggestions on the price of their choices, making effectivity a built-in design precept, not an afterthought.
This displays a broader shift towards AI-native apps constructed on globally distributed, serverless knowledge platforms like Azure Cosmos DB.
Walmart: Reliability and efficiency at scale
Whereas AI is remodeling how functions are constructed, one factor hasn’t modified: Efficiency and reliability stay mission-critical.
As Kirill emphasised, AI doesn’t take away the necessity for reliability, safety, and efficiency.
In truth, it raises the bar. This was bolstered in periods like Walmart’s, the place Technical Fellow Sid Anand defined that large-scale functions should:
- Ship low-latency experiences globally.
- Stay out there via regional failures.
- Keep constant efficiency at huge scale.
“We would like folks to have the ability to add to their cart and look at cart it doesn’t matter what is occurring in a given cloud area…and we’d like all of those interactions to be low latency as a result of any kind of latency friction will trigger a drop-off,” stated Sid.
From gigabytes to petabytes, from lots of to trillions of transactions, trendy programs should function seamlessly below unpredictable demand.
These necessities align with how Azure Cosmos DB is designed for international distribution and low latency at scale.
Value effectivity turns into a core design precept
A remaining takeaway from Cosmos Conf: as programs develop extra complicated, price turns into simply as essential as scale.
Throughout the keynote and periods, we noticed a transparent shift:
- Builders want price visibility in actual time.
- Architects have to design for effectivity upfront.
- Groups need to consolidate platforms and cut back complexity.
That is the place improvements like Azure DocumentDB come into focus.
As highlighted within the keynote, Azure DocumentDB affords over 40% decrease price vs. alternate options, and permits excessive efficiency with simplified structure. It additionally helps open-source, multi-cloud portability eventualities. The result’s a broader alternative for builders:
Design and structure examples that builders can begin constructing now
Past the keynote, there have been numerous demo-driven periods at Cosmos Conf throughout app architectures, repeatable patterns, and finest practices for constructing and scaling AI-enabled options.
For instance, Farah Abdou, a lead machine engineer at startup SmartServe, shared how her group rebuilt their structure utilizing Azure Cosmos DB as a unified “agent reminiscence material.” By combining vector seek for semantic caching, change feed for event-driven coordination, and optimistic concurrency for battle prevention, they have been in a position to cut back prices, allow sub-100ms agent handoffs, and remove state conflicts.
One other subject we get requested about quite a bit is how customers shield and govern their AI functions. Pamela Fox, a Microsoft Principal Cloud Advocate, walked via how one can construct safe, multi-user AI programs utilizing the Mannequin Context Protocol (MCP). By authenticating customers with Entra ID and storing per-user knowledge in Azure Cosmos DB, she enabled role-based entry with Microsoft Graph, and sensible growth workflows utilizing instruments like VS Code and GitHub Copilot.
From these hands-on patterns to large-scale manufacturing programs, the lesson was constant: groups are designing for scale, effectivity, and real-world utilization from day one.
Key takeaways
- AI functions require versatile, schema-agnostic knowledge fashions.
- Serverless and immediate scalability have gotten default expectations.
- Semantic and vector search are actually core to utility design.
- Value visibility and effectivity should be designed upfront.
Constructing for what’s subsequent
We’re coming into a brand new period of utility growth. Apps have gotten AI-native, globally distributed, and are repeatedly evolving.
And success will rely upon how nicely organizations align to those shifts.
Essentially the most forward-thinking groups we heard from at Cosmos Conf are already doing this by:
- Designing for flexibility.
- Constructing for velocity, not simply scale.
- Treating price and efficiency as key considerations.
- Leveraging AI not simply in apps, however in how apps are constructed.
This isn’t only a expertise shift.
It’s a shift in how we take into consideration constructing software program.
Discover Cosmos DB Conf on demand
Should you missed Cosmos Conf 2026, you’ll be able to discover all periods on demand and listen to straight from the groups constructing these programs in manufacturing in the present day.
The patterns shared this 12 months are greater than finest practices, they’re a blueprint for what comes subsequent.