Cypher Agency

Most Australian organisations know exactly where their core data platform lives. Almost none can answer the same question for AI. We're publishing a piece this week on what actually happens the moment a prompt leaves someone's laptop - and why the answer matters more than most governance frameworks currently account for. More to come.

"AI Engineer" is one of the fastest-growing job titles of the past two years, and it means something different at almost every organisation using it. We've written a working definition: AI engineering sits at the intersection of data science, data engineering, and software engineering — overlapping meaningfully with each, reducible to none of them. Two skill sets most job descriptions blur together: the technical side (model integration, retrieval, evaluation, cost engineering) and the software engineering side most consistently underweighted (solution design, integration, production fundamentals, governance literacy). Just as important: where AI engineering's responsibility ends. We draw the boundary against data engineering, DevOps, and security explicitly, rather than treating all three as the same job. Companion piece to our Data Team 2026 argument. Full guide in the first comment. 👇 #AIEngineering #DataLeadership #DataIntegration #CypherAgency

A $600,000 Salesforce contract, cancelled in two months. Enterprise transformations that run for years, weighed down by governance nobody can shortcut. Keith Jenneke breaks down why both are true at once — and why the difference isn't AI capability, it's shape. A workflow small enough for one team to own end to end is where build now beats buy. An enterprise transformation buried in compliance and financial controls is a different problem, and AI-assisted coding doesn't touch it. The part getting less attention than the dollar figures: a lot of this new software is going up with no governance at all. Full piece in the first comment. 👇

Most orgs still argue over who "owns" AI. The better question: does anyone own the whole data path? Our latest piece breaks down what a converged Data Team 2026 actually looks like — five roles, one shared platform, one governance model, no gaps between integration, analytics, and AI. #DataTeam2026 #DataIntegration #DataLeadership #AIGovernance #CypherAgency

Converged Data Teams in 2026. Three separate teams. Three roadmaps. Three sets of assumptions about what the data actually means. That split made sense when integration, analytics, and AI each needed a genuinely distinct toolchain. It's starting to cost more than it saves. The symptom: an integration team makes a schema call upstream. A data team models on top of it, unaware. An AI initiative builds on both, with no visibility into whether any of it was ever validated. Every decision is reasonable alone. Together, they produce exactly the inconsistent, hard-to-trust platform that undermines confidence in whatever AI sits on top. Not an argument for one undifferentiated role — deep integration and AI expertise stay genuinely distinct. It's an argument for one roadmap, one governance model, with specialisation living inside that structure instead of scattered across three disconnected ones. We've written up what that looks like in practice, and the one diagnostic question worth asking about your own team's structure. Link in the first comment. #DataLeadership #DataTeam2026 #DataIntegration #CypherAgency

Episode 008 of Data Engineer Comics: a "small" CSV upload — one file, a few columns, needed by end of day — brought down 17 pipelines and every downstream dashboard. The file: sales_update_FINAL.csv. 1,248,394 rows. No schema contract, no data types, no validation. A hidden column, dates in seven different formats, business logic buried in Excel before the file ever reached the pipeline. Root cause: optimism bias, no safeguards, and no one saying no to a request that should have triggered a schema conversation before it triggered an upload. The fix that should have existed from the start: a schema contract, validation at the door, standardised parsing, and quarantine for anything that doesn't conform — the same discipline behind our Autoloader ingestion pattern. A CSV is not a format. It's a lifestyle. If your pipeline depends on someone's desktop file, you have a time bomb, not a pipeline. #DataEngineering #DataIntegration #DataQuality #CypherAgency

78% of developers say they're coding faster with AI. 79% say their organisation's overall software delivery hasn't gotten any faster. GitLab's 2026 AI Accountability Report calls it the AI Paradox. The bottleneck didn't disappear — it moved downstream, to review and validation, which hasn't scaled at the same rate. 85% agree AI has shifted the constraint from writing code to reviewing it. 34% of organisations that had a production incident involving AI-generated code couldn't actually determine whether AI-generated code caused it. This applies directly to data platform work, arguably with sharper consequences. A broken pipeline doesn't fail loudly — it can run for weeks producing plausible, wrong output before anyone notices. The response isn't less AI-assisted development. It's building review and governance into the platform itself, not bolting it on after. We've written up what this means for teams building Data Integration, Analytics, and AI platforms. #AIGovernance #DataEngineering #SoftwareDelivery #CypherAgency

Episode 007 of Data Engineer Comics illustrates a genuine risk within any Data Integration, Analytics, and AI environment: an untracked platform, built as a temporary fix three years ago, discovered to be quietly powering twenty-three downstream processes at a monthly cost of $2,920 — entirely outside governance, monitoring, or backup coverage. The root cause is a familiar one. Infrastructure is easy to create and easy to forget when no discovery process exists to catch it. The lesson: visibility is a security control. For any organisation building genuine Data Integration, Analytics, and AI capability, discovery controls and clear ownership are not optional — they determine whether an environment can be trusted. #DataIntegration #DataGovernance #DataAnalyticsAndAI #CloudGovernance

As of 6 July 2026, Databricks Genie operates under pay-as-you-go pricing. Free monthly allowance per user, then billed on usage beyond that. This reaches two audiences differently. Managers: are Unity AI Gateway budgets actually configured, is Genie Space curation owned, is there a defined escalation path for alerts. Practitioners — data engineers, analysts, data scientists: are you using Genie selectively, or defaulting to it for everything? Reserve it for genuinely exploratory questions. Save recurring reports as dashboards instead of re-asking. Don't use it to preview the first ten rows of a table. Batch related questions instead of iterating through five narrow ones. Governance without practitioner awareness produces budget alerts that surprise the people who triggered them. Practitioner discipline without governance produces good individual habits with no organisational visibility. Neither works alone. Full checklist for both audiences in the article — link in the first comment.

"Just updating a tag and a security group." The plan comes back: 113 to add, 482 to change, 872 to destroy. Episode 006 of Data Engineer Comics. The cause: an S3 bucket ACL quietly changed via console click-ops three months ago. Terraform's state file never knew. The plan isn't wrong — it's just brutally honest about everything nobody documented. State is a contract. Console changes have a cost. Automation without governance is just faster chaos. Until someone reviews the plan. #Terraform #InfrastructureAsCode #CloudGovernance #CypherAgency