The company launched agentic video understanding on 1 September 2026 for Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.5 Flash-Lite.

It is available for uploaded videos and YouTube videos through the Gemini API in Google AI Studio and Google’s Gemini Enterprise Agent Platform.

Google says its testing found that the new approach can use up to 88% fewer tokens, reduce analysis costs by up to 66%, and improve accuracy by up to 7% compared with conventional video processing.

Those are Google’s benchmark results rather than a promise that every video workload will achieve the same savings.

Gemini does not need to watch every second in the same way

The important change is how Gemini processes video.

By default, conventional Gemini video processing samples footage at a fixed rate — normally one frame per second — and puts that information into the model’s context.

That approach is relatively straightforward, but it means longer videos consume increasingly large numbers of tokens even if only a small section of the footage is relevant to the user’s question.

Agentic video understanding changes this.

Gemini can dynamically navigate through the video’s timeline and decide whether it needs to inspect frames, audio or the transcript.

It can also alter how closely it examines a particular section.

For example, the model could scan through a long recording relatively quickly, identify a potentially important moment and then revisit that short section at a higher frame rate.

Google describes the system as an agentic loop in which Gemini can repeatedly load the parts of the video it needs before producing its answer.

That distinction is particularly important for long videos.

Why fewer tokens matter

AI models do not see an hour-long video in quite the same way a human simply watches it.

Video frames, audio and other information need to be represented as tokens that the model can process.

Google’s documentation says static video processing can use approximately 100 tokens per second at the default low media resolution, or around 300 tokens per second at higher resolution.

That can become substantial over a long recording.

Google gives an illustrative example in its developer documentation where one hour of video could consume roughly 1.08 million tokens using static processing, compared with around 108,000 using agentic processing in the example scenario.

Actual consumption depends on the video and question being asked.

The advantage is therefore not simply that Gemini can analyse video differently.

It potentially avoids paying the computational cost of examining information that is irrelevant to the question.

Google says Gemini can find moments lasting less than a second

The ability to revisit relevant sections also allows Gemini to deal with events that conventional one-frame-per-second sampling might miss.

Google lists sub-second moment retrieval as one of the feature’s main uses.

It could, for example, help identify the precise moment at which something changes in footage rather than returning only a rough timestamp.

Google also highlights anomaly detection.

If Gemini notices something unusual in a particular time window, it can inspect that section at a higher frame rate to look for rapid movement or visual details that might have been missed during the initial scan.

Other stated applications include counting repeated actions or objects and searching for specific information inside hours of video.

Long videos are where the change becomes more interesting

The technology makes considerably more sense when dealing with a two-hour lecture or hours of security footage than when analysing a 20-second video clip.

Google says its efficiency improvements were particularly pronounced on longer content, ranging from 10-minute instructional videos to 90-minute lectures and multi-hour recordings.

For a straightforward question about a short video, the additional agentic navigation can actually introduce some extra time before the model starts answering.

Google’s developer documentation warns that agentic processing may slightly increase time to first token for short clips under five minutes because the model first has to reason about how it will navigate the footage.

That makes the technology less about replacing every existing video-processing method and more about choosing a smarter method for complicated or lengthy footage.

Why it matters in South Africa

Unlike some AI launches that arrive with unclear regional availability, the underlying developer services are available locally.

Google officially lists South Africa as a supported country for both Google AI Studio and the Gemini API.

That gives the technology possible applications well beyond ordinary consumer video search.

South African companies increasingly generate large amounts of footage across security systems, broadcasting, retail, mining, manufacturing and training environments.

Finding a particular event across hours of recordings can be expensive if an AI system has to process every frame at the same level of detail.

A model that first searches broadly and only inspects relevant sections more closely could make some of those workloads less computationally intensive.

Security-camera analysis is an obvious example, although any organisation processing surveillance or other sensitive footage would still need to assess privacy, data-governance and regulatory requirements before sending video to an AI service.

The same technology could also be useful for broadcasters searching archive footage, education providers indexing recorded lectures, and companies analysing training or industrial inspection video.

The feature is coming to ordinary Gemini users too

Agentic video understanding is initially a developer and enterprise feature, but Google does not intend to keep it there.

The company says the technology will roll out to users of the Gemini app across its Flash and Flash-Lite models.

Google also plans to use the technology for Ask YouTube, allowing the feature to provide answers based more accurately on the visual contents of videos.

Google has not provided an exact date for those consumer rollouts, saying Gemini app support is coming soon and Ask YouTube integration will follow in the coming months.

For developers, however, agentic video understanding is available now.

And the bigger change may be conceptual: Gemini is moving from simply accepting a video as input to actively deciding how it needs to watch it.

The post Google Gemini can now decide which parts of a video to watch — using up to 88% fewer tokens appeared first on Technology Blog.

Oracle made Database 23ai generally available on 2 May 2024 as its next long-term support database release. The company had originally called the release Oracle Database 23c before renaming it 23ai to reflect its increased emphasis on artificial intelligence.

There is an important update for anyone researching the product today. Oracle Database 23ai is no longer Oracle’s current long-term database name. Oracle introduced Oracle AI Database 26ai on 14 October 2025, replacing Database 23ai while carrying its major technologies forward.

Existing 23ai customers did not need a conventional database upgrade to make that transition. Oracle says customers could apply the October 2025 Release Update to move from 23ai to 26ai without application recertification.

This means Database 23ai remains important because it established much of the technical foundation now found in Oracle AI Database 26ai.

What was Oracle Database 23ai?

Oracle Database 23ai was Oracle’s enterprise relational database platform with more than 300 new features centred on artificial intelligence, developer productivity, application development, security and mission-critical workloads.

It retained Oracle Database’s traditional relational database capabilities while introducing ways to work directly with vectors, JSON documents, property graphs and newer application-development patterns.

The strategy was not to turn Oracle Database into a database dedicated only to AI. Instead, Oracle added AI data types and search capabilities alongside existing relational, JSON, spatial, graph and transactional data.

That distinction is central to understanding Database 23ai.

Businesses could keep operational information in Oracle Database while adding semantic search or retrieval-augmented generation functionality without necessarily moving that information into a separate dedicated vector database.

AI Vector Search was one of 23ai’s biggest additions

Oracle AI Vector Search was among the headline technologies introduced with Database 23ai.

Traditional database searches often rely on exact values, filters or keywords. Vector search instead represents information mathematically as vectors, allowing an application to search according to similarity or meaning.

This can be useful with unstructured information such as:

• documents;
• product descriptions;
• images;
• audio;
• support information;
• knowledge bases;
• text embeddings.

Database 23ai introduced a native VECTOR data type, allowing these embeddings to sit alongside conventional business information inside database tables.

Developers could then combine vector similarity searches with relational, JSON, graph, spatial and text criteria using SQL.

For example, an enterprise application could search a collection of support documents semantically while also applying normal business rules such as customer identity, product ownership, region or access permissions.

Vector search gives Oracle a foundation for RAG applications

One important use of vector search is retrieval-augmented generation, commonly shortened to RAG.

RAG allows an application to find relevant information from a controlled data source and provide that material as context to a large language model.

This can help an AI application answer questions using an organisation’s own information instead of relying only on the model’s original training data. Oracle specifically positioned AI Vector Search for enterprise RAG applications involving private business information.

Possible applications include an internal assistant that searches company procedures, a customer-service tool that retrieves product documentation or an application that finds information across technical records.

RAG does not guarantee an accurate AI response. Organisations still need to consider source quality, permissions, model behaviour, prompt design and how generated answers are validated.

Vector search can stay close to existing business data

One of Oracle’s arguments for integrating vector search into its database is that organisations may not need to maintain a separate vector platform alongside their transactional database.

Oracle allows vector queries to interact with conventional relational information through SQL.

This architecture can reduce the need to copy selected enterprise data into another specialised database purely so that an AI application can perform semantic searches.

It can also simplify the application architecture where an organisation already relies extensively on Oracle Database.

However, that does not mean every AI workload should automatically run inside an Oracle database. Infrastructure cost, existing platforms, development skills, application architecture and required model support remain important considerations.

JSON Relational Duality gives developers two views of the same data

Another major Database 23ai feature was JSON Relational Duality.

Traditional relational databases organise information into tables, rows and columns. Modern application developers often prefer working with JSON documents because these can map more naturally to application objects.

JSON Relational Duality attempts to combine both approaches.

Oracle can store information using its relational model while presenting that information to an application as JSON documents through Duality Views. Changes can also flow back through those views without requiring developers to maintain duplicate relational and document copies of the same information.

This can be useful when a database administrator wants the consistency and structure of relational storage while application developers want a document-oriented programming model.

Oracle describes the approach as a way of reducing the compromises that traditionally come from choosing either relational storage or a document database.

Operational property graphs make relationships easier to analyse

Database 23ai also expanded Oracle’s support for property graphs.

Graph analysis focuses on relationships between information rather than looking only at individual records.

Potential use cases include:

• detecting relationships between financial transactions;
• analysing supply chains;
• mapping customer relationships;
• fraud investigation;
• network analysis;
• recommendation systems;
• understanding connected assets.

Oracle allows developers to perform operational property-graph queries against data already stored in the database and supports the SQL Property Graph Queries standard, SQL/PGQ.

This creates another example of Oracle’s converged database approach: relational, JSON, vector and graph workloads can operate against related enterprise data rather than automatically requiring separate platforms for each data model.

Database 23ai made several SQL changes for developers

Not every Database 23ai improvement involved AI.

Oracle also made numerous changes intended to make SQL and application development more straightforward.

One significant addition was a native BOOLEAN data type in SQL, giving developers direct true-and-false values rather than requiring alternative representations.

Other changes included:

• optional FROM clauses in applicable SQL statements;
• IF NOT EXISTS support for selected DDL operations;
• multi-value inserts;
• table value constructors;
• aliases in GROUP BY;
• additional RETURNING functionality;
• joins in UPDATE and DELETE;
• SQL domains;
• database object annotations.

These changes are less headline-grabbing than generative AI, but they can affect everyday application development more frequently.

True Cache addresses read-heavy application workloads

Oracle True Cache was another notable technology associated with Database 23ai.

True Cache acts as a read-only database cache in front of the primary Oracle Database. Frequently accessed information can be served from the cache while the primary database continues handling authoritative data and write operations.

Oracle automatically keeps the cache synchronised with changes to the underlying database.

This can reduce query load on the main database and improve response times for applications that perform many more reads than writes.

An online booking application provides a straightforward example. Users may repeatedly search available products or services, generating large numbers of read operations, while comparatively fewer customers complete transactions that change database records.

True Cache can serve appropriate read requests while the primary database remains responsible for transactional changes.

Lock-Free Reservations target high-volume transactions

Database 23ai also introduced Lock-Free Reservations.

In a conventional transactional database, changing a value can require locking a row while the transaction completes. Under heavy concurrency, competing transactions may then have to wait.

Lock-Free Reservations allow applications to reserve part of a numeric value without locking the entire row. Oracle gives examples such as reserving part of an account balance.

Potential use cases include systems dealing with:

• financial balances;
• inventory;
• ticket availability;
• resource allocations;
• reservations;
• other highly concurrent transactions.

The objective is to reduce contention when many transactions attempt to modify the same resources.

SQL Firewall adds protection inside the database

Security was another major focus.

Oracle Database 23ai introduced SQL Firewall, which can observe normal SQL activity and create an allow list of authorised SQL patterns.

Administrators can then configure the database to log or block SQL that falls outside those approved patterns.

This is intended to provide another line of defence against threats including SQL injection and unauthorised SQL execution.

A database firewall does not replace secure application development.

Businesses still need proper authentication, access control, application patching, network security, credential management, monitoring and vulnerability management.

Where businesses could use Oracle Database 23ai

The combination of traditional database features and newer AI capabilities gave Database 23ai several potential enterprise use cases.

Enterprise AI search

A company could embed documents and other information into vectors and build semantic search across internal knowledge.

Retrieval-augmented generation

Applications could retrieve relevant private business information before sending context to an LLM.

Customer-service systems

AI Vector Search could help find relevant support documents, product information or historical knowledge while ordinary database queries enforce customer and business rules.

Financial applications

Transactional features, graph analysis and SQL Firewall can support systems where data integrity, relationships and database security matter.

ERP and operational applications

Organisations running custom operational or ERP-style systems can use Oracle’s relational database functionality while modernising application interfaces with JSON or adding vector search to selected datasets.

Fraud and relationship analysis

Property graphs can analyse relationships between accounts, transactions, devices or other entities.

E-commerce and reservation systems

True Cache and Lock-Free Reservations address workloads involving large numbers of reads or competing transactions.

Application modernisation

JSON Relational Duality can allow developers to work with JSON documents while retaining relational storage underneath.

Oracle Database 23ai features at a glance

Oracle’s 23ai documentation listed more than 300 new features in total, so these represent the major platform changes rather than an exhaustive feature list.

Oracle Database 23ai has now become Oracle AI Database 26ai

This is the most important fact for businesses evaluating Database 23ai in 2026.

On 14 October 2025, Oracle introduced Oracle AI Database 26ai as its next AI-focused database release and replacement for Database 23ai.

The change was unusual because Oracle did not require customers already running 23ai to perform a conventional major database upgrade.

Oracle says customers could transition by applying the October 2025 Release Update, with no database upgrade or application recertification required.

Oracle subsequently updated its product documentation to use the Oracle AI Database name.

As of 30 August 2026, organisations researching a new Oracle deployment should therefore investigate Oracle AI Database 26ai, not plan a new installation around 23ai as though it remained the latest release.

26ai builds on the technologies introduced with 23ai

The rename and version change did not discard the 23ai technology.

AI Vector Search, JSON Relational Duality, graph support, True Cache and other capabilities continue into Oracle AI Database 26ai.

Oracle has also expanded its AI strategy further.

Current 26ai capabilities and announced functionality include deeper AI integration, agentic AI workflows, Model Context Protocol support, Apache Iceberg integration and additional ways to use LLMs with enterprise information.

This makes 23ai useful historically as the release where Oracle’s current AI-database direction became much more explicit.

South African organisations can use Oracle’s Johannesburg cloud region

Oracle also has local cloud infrastructure relevant to South African organisations.

The OCI South Africa Central region in Johannesburg, identified as af-johannesburg-1, has operated since January 2022 and remains listed as a live Oracle Cloud Infrastructure region.

Oracle also lists an Interconnect for Microsoft Azure between its Johannesburg OCI region and Azure South Africa North.

A local cloud region can be important where latency, application architecture or data-location requirements influence infrastructure decisions.

However, the existence of a South African OCI region should not be interpreted as proof that every Oracle database service, configuration or feature is available locally. Oracle notes that cloud-service and SKU availability can differ between regions.

South African businesses working with personal information must also evaluate their own obligations under POPIA. Choosing a local cloud region does not by itself make a database deployment POPIA compliant.

Enterprise AI requires more than adding vectors to a database

Database 23ai arrived during a period when businesses increasingly wanted generative AI systems to interact with internal corporate information.

Oracle’s answer was to bring vector search and AI functionality closer to the operational data already held in its databases.

That architecture can remove some data movement and reduce the number of specialised systems an organisation needs to manage. However, database functionality only solves part of an enterprise AI project.

Businesses still need policies around data access, model selection, information quality, human oversight and the actions AI systems are permitted to perform.

TechnologyBlog.co.za has previously examined some of the organisational questions around introducing AI into company decision-making in AI in the Boardroom. Read AI in the Boardroom on TechnologyBlog.co.za

The issue becomes even more significant as databases move beyond information retrieval towards AI agents capable of interacting with enterprise systems.

Should a business still deploy Oracle Database 23ai?

For a completely new deployment in 2026, 23ai should no longer be treated as Oracle’s latest database release.

Oracle AI Database 26ai has replaced it as the current long-term release, and Oracle’s current documentation, free downloads and product messaging now point new users towards 26ai.

Existing 23ai installations are different.

Because Oracle designed the transition from 23ai to 26ai around a Release Update rather than a conventional database upgrade, organisations already using 23ai should review Oracle’s current update documentation and their own application-support requirements before deciding on an upgrade schedule.

The technologies introduced with Database 23ai remain highly relevant. AI Vector Search, JSON Relational Duality, graph queries, True Cache and its SQL improvements continue to form part of Oracle’s current database platform.

The product name has moved on. Much of the architecture has not.

The post Oracle Database 23ai brought AI Vector Search to enterprise data — but it is now Oracle AI Database 26ai appeared first on Technology Blog.