Practitioner's Corner
Lessons from the field—what we see building at scale
Practitioner's Corner
Lessons from the field—what we see building at scale
The Six-Hour Price Check
An analyst opens their laptop at 9 AM. The task: check competitor pricing across five major players in three product categories. Document changes. Flag anything significant. Seems straightforward. By 3 PM, they're still at it. This is "just check the website" in practice. Six hours navigating to pricing pages, documenting numbers in spreadsheets, cross-referencing yesterday's data. Repeat tomorrow. And the day after.
The Six-Hour Price Check
An analyst opens their laptop at 9 AM. The task: check competitor pricing across five major players in three product categories. Document changes. Flag anything significant. Seems straightforward. By 3 PM, they're still at it. This is "just check the website" in practice. Six hours navigating to pricing pages, documenting numbers in spreadsheets, cross-referencing yesterday's data. Repeat tomorrow. And the day after.
The Oncologist Who Builds AI for Two-Minute Cancer Decisions
Two minutes. That's how long most UK tumor boards have to discuss each patient's cancer treatment plan. In that window, clinicians synthesize radiology reports, pathology results, biomarker tests, and treatment guidelines to make decisions that directly affect survival. When information gaps appear—missing test results, incomplete staging—the case gets postponed. This happens in 7% of cases.
Dr. Andrew Soltan saw this from inside. He's a practicing oncologist at Oxford University Hospitals who decided to build AI agents for his own workflow. When someone who actually makes these two-minute decisions designs the system meant to support them, the constraints look different. And what "production-ready" means becomes something you can measure in survival outcomes.
The Oncologist Who Builds AI for Two-Minute Cancer Decisions
Two minutes. That's how long most UK tumor boards have to discuss each patient's cancer treatment plan. In that window, clinicians synthesize radiology reports, pathology results, biomarker tests, and treatment guidelines to make decisions that directly affect survival. When information gaps appear—missing test results, incomplete staging—the case gets postponed. This happens in 7% of cases.
Dr. Andrew Soltan saw this from inside. He's a practicing oncologist at Oxford University Hospitals who decided to build AI agents for his own workflow. When someone who actually makes these two-minute decisions designs the system meant to support them, the constraints look different. And what "production-ready" means becomes something you can measure in survival outcomes.
Rina Takahashi
Rina Takahashi, 37, former marketplace operations engineer turned enterprise AI writer. Built and maintained web-facing automations at scale for travel and e-commerce platforms. Now writes about reliable web agents, observability, and production-grade AI infrastructure at TinyFish.
The Number That Matters
Data professionals spend 40% of their time on quality tasks, with freshness violations causing nearly a third of all data downtime. Engineers end up running constant fire drills instead of building new pipelines.
The freshness problem shifts wildly by context. Fraud detection algorithms need sub-second latency. Marketing dashboards can wait a week. The same extraction system serves both, each with its own decay curve and operational burden.
What gets logged as "successful extraction" often hides the real work: validating timestamps, flagging stale records, managing refresh schedules across dozens of sources that age at different rates.
Data professionals spend 40% of their time on quality tasks, with freshness violations causing nearly a third of all data downtime. Engineers end up running constant fire drills instead of building new pipelines.
The freshness problem shifts wildly by context. Fraud detection algorithms need sub-second latency. Marketing dashboards can wait a week. The same extraction system serves both, each with its own decay curve and operational burden.
What gets logged as "successful extraction" often hides the real work: validating timestamps, flagging stale records, managing refresh schedules across dozens of sources that age at different rates.
Freshness equals time since most recent timestamp. If latest transaction is two days and forty-five minutes old, freshness reads 2.03 days.
Financial time-series data loses value in milliseconds. Customer demographic records stay relevant for months despite individual changes happening constantly.
Real-time marketing needs minute-level web analytics. Trend analysis tolerates much longer intervals. Same data, radically different freshness requirements.
Stock prices and social feeds go stale fast, but nobody notices until downstream systems flag anomalies or business decisions fail.
Teams spend time on reactive freshness fixes and quality fire drills instead of building valuable new pipelines or extracting more insight.
Field Notes from the Ecosystem
December delivered the kind of operational lessons that only surface when systems run at scale. Compromised services hijacked entire fleets. Configuration changes cascaded into global outages. The gap between compliant bots and evasive scrapers that outnumber them eight to one became impossible to ignore.
We also saw what production optimization actually looks like in numbers. Latency dropped fivefold. Uptime percentages slipped just enough to matter. These observations come from infrastructure meeting real workload, from authentication protocols getting deployed, from the distance between pilot programs and production systems becoming measurable in downtime minutes.
December delivered the kind of operational lessons that only surface when systems run at scale. Compromised services hijacked entire fleets. Configuration changes cascaded into global outages. The gap between compliant bots and evasive scrapers that outnumber them eight to one became impossible to ignore.
We also saw what production optimization actually looks like in numbers. Latency dropped fivefold. Uptime percentages slipped just enough to matter. These observations come from infrastructure meeting real workload, from authentication protocols getting deployed, from the distance between pilot programs and production systems becoming measurable in downtime minutes.
Practitioner Resources
