Cobnex blogs / Australia

Cobnex blog

AI implementation, software architecture, cloud operations and Australian technology policy.

Australian AI / Policy briefings

Australian AI policy5 min read

Australia's AI direction still leaves the hard project choices to you

The National AI Plan sets a national direction. Turning that direction into a useful business system requires a narrower task, reliable data and evidence that the proposed workflow improves on today's process.

Australian AI governance5 min read

Government AI assessments need evidence before the deadline

The DTA policy separates new and existing use cases. For delivery teams, the practical work is to establish scope and connect the assessment to the system's real data flows, permissions and tested behaviour.

512 articles

Page 1 of 29

Permission aware retrieval4 min read

Search results must respect the person asking

A knowledge assistant needs the same access boundaries as the documents behind it. Here is how to place those checks without making the model responsible for security.

Hybrid knowledge search3 min read

When keyword search beats a vector match

Product codes and everyday questions need different retrieval behaviour. Hybrid search combines exact wording with meaning, but the combination needs a clear reason to exist.

Document chunk boundaries4 min read

A table cell is not a complete answer

Splitting a document into searchable passages can remove the information that makes a sentence or number meaningful. Choose boundaries around the questions the source needs to answer.

Grounded answer citations3 min read

A citation can be real and still support the wrong claim

Linking to a genuine document does not prove an answer is correct. The evidence must support the particular claim, including its conditions and exceptions.

Retrieval freshness3 min read

Deleting a policy should also retire its cached answers

Removing a document from search does not remove the answers already built from it. Design freshness around the complete route from source to reader.

Prompt injection boundaries3 min read

Supplier documents must not become agent instructions

An assistant may need to read an external document without accepting its requests. Keep the authority to act in application code and authenticated user intent.

Bounded agent tools3 min read

Give an agent a small tool, not an administrator console

A useful agent capability describes a business operation with clear limits. General administrative access makes mistakes harder to contain and results harder to explain.

Human approval checkpoints3 min read

An approval should authorise one exact proposal

A reviewer needs to approve the action that will actually run. Tie the decision to a saved target, payload and version so later changes cannot inherit it.

Agent recovery checkpoints3 min read

Resuming an agent is more than replaying its conversation

Conversation history explains intent. Durable operation records establish what happened. Recovery needs both, with a clear rule for uncertain side effects.

AI evaluation datasets3 min read

A better demo is not an evaluation dataset

A handful of polished examples can show what an assistant might do. A useful evaluation set shows where it works, where it fails and what those failures mean.

Model change releases3 min read

Release a model change like a behaviour change

A replacement model can keep the same API shape while changing answers, tool choices and refusals. Assess those behaviours as part of the application release.

Document intelligence validation3 min read

Invoice extraction needs an arithmetic check

Reading the right characters is only the first step. Before accepting an invoice, check that the selected fields mean the right things and agree with one another.

Private inference boundaries3 min read

Private inference has more than one data boundary

A private network connection addresses one part of an AI system. Prompts, retrieval stores, diagnostics and support access still need their own data-handling decisions.

AI task cost budgets3 min read

Price the completed AI task, including its retries

A task budget should cover the whole attempt to produce a useful result. Model calls, tools, validation and recovery all draw from that budget.

AI abstention design3 min read

Knowing when an assistant should decline to answer

Missing evidence, unclear intent and missing authority need different responses. Give the product a useful path for each instead of one generic refusal.

Modular monolith boundaries3 min read

A modular monolith needs boundaries the code can enforce

One deployable application can contain clear business modules. The separation becomes useful when callers cannot casually bypass each module's rules and data ownership.