A database transaction is a very small drama. Something changes: a price, a balance, a seat, a shipment. The old value exits; the new one enters. For most of computing history, companies treated the moment itself as an administrative detail. Capture it, store it, inspect it later. Steve Wilkes built a career around the suspicion that “later” was doing rather too much work.
At GoldenGate Software, where Wilkes led an advanced technology group, the company’s specialty was moving database transactions from one system to another. Customers regularly posed an awkwardly sensible question. If the information was already in motion, could they look at it then? Could they decide something before it reached its destination and cooled into history?
The question followed him through Oracle’s 2009 acquisition of GoldenGate. Wilkes continued his data-integration work and took the lead on Oracle’s cloud data-integration strategy. In 2012, he and three former GoldenGate colleagues founded what became Striim. They did not need to invent a customer problem. They had been listening to one for years.
The first name was WebAction, a label with the brisk confidence of a browser button. The company later became Striim, adding two letters to the obvious word and a permanent obligation to explain the pronunciation. Behind the new spelling was a widening ambition. Copying changes between databases was only the beginning. The team wanted to combine transaction streams with logs, messages and sensor readings; process them in memory; find patterns; and send a useful result onward without demanding a rest on disk. The founders had already seen the installed systems, the awkward edge cases and the difference between a demonstration that runs and a pipeline that survives Monday morning.
“Data is not created in batches.”Steve Wilkes, on the natural rhythm of information
01 / The useful secondA business can be right and still be late
Wilkes’s argument for streaming data begins in ordinary life. Imagine learning about a traffic accident after arriving home from the jam. The report may be perfectly accurate. It is also useless. Many business facts have the same brief shelf life: a payment that looks suspicious, an aircraft part drifting out of tolerance, an online basket left idle, a shipment about to miss its connection. Their value depends less on permanent truth than timely arrival.
Batch processing was a practical bargain made when storage was cheap and computing power was dear. Accumulate the records, run the job at night, return in the morning to see what happened. The bargain hardened into habit. Wilkes’s preferred unit is the event - a row, a line, a signal - handled in the shape and sequence in which the world produced it.
That middle box is where Striim lives. Its work joins change data capture, continuous processing, delivery and monitoring. The company name is pronounced “stream,” a small mercy in a sector that often makes pronunciation the first technical challenge. Wilkes has spent years teaching the model through webinars and tutorials, often using a SQL-like language to show that continuous processing need not require an entirely alien grammar.
In interviews, he made the uses deliberately plain. An online retailer can notice that a basket has been abandoned and offer an incentive while the shopper is still reachable. A stream of device readings can be joined with context before an abnormal pattern becomes a failure. Database changes can be transformed on their way to an analytics system instead of becoming another cleanup queue at the destination. The examples share a structure: live information, a narrow window and a response whose value declines with every minute. “Real time” is an elastic phrase in marketing. In these cases, the decision supplies the honest clock.
02 / An engineer in the executive suiteThe title grew; the workbench stayed
There is a tidy corporate version of Wilkes’s biography: a Master of Engineering in microelectronics and software engineering from Newcastle; technology and product roles at Cap Gemini, AltoWeb and The Middleware Company; GoldenGate; Oracle; founder and CTO. The more revealing version is untidier. He has worked across analysis, architecture, development, testing, management, sales and marketing. The connective skill is translation - between machines, certainly, but also between the people who buy software and those who build it.
Former colleagues describe a patient mentor who gave engineers room to explore and could move easily from a customer request to its technical feasibility. That range helps explain the public demos. Wilkes does not merely announce that streams can be enriched. He walks through the continuous query, joins the live data to a cache and inspects the resulting records. It is difficult to float too far into executive vapor while the code is still compiling on your screen.
His engineering record includes work on real-time transactional data obfuscation, co-invented with Shenoda Guirguis and Alok Pareek. The patent family addresses a delicate problem: obscure sensitive information while transactions move, preserve enough statistical character for analysis, and avoid parking an exposed copy at a third-party destination first. Privacy, in this formulation, is not a cleaning job performed after delivery. It belongs inside the journey.
The design is an instructive piece of Wilkes’s larger theme. A copied banking transaction might be useful for fraud analysis, yet copying the original sensitive fields to an outside environment before masking them creates the very exposure the exercise should prevent. The proposed methods select obfuscation techniques by data type and apply them during replication. Referential integrity and analytical usefulness have to survive even when the original values do not. Speed and governance, so often introduced as rivals in sales presentations, become constraints on the same engineering problem.
“I still code in multiple languages, often at the same time.”Steve Wilkes, describing his hands-on role
03 / The bill behind the billPlumbing acquires furniture
Wilkes’s recent writing returns to the systems he knew at GoldenGate, now seen from the other end of two decades. A legacy replication engine may still copy rows competently. Around it, however, an organization can accumulate a small suburb of compensating machinery: external cluster managers, transformation jobs, masking rules, validation scripts, scheduled checks and the human runbooks that keep them civil.
His suggestion is disarmingly practical. At renewal time, list every tool, script and scheduled process that exists because the replication engine cannot do the surrounding work. Then ask which pieces the team genuinely wishes to keep operating. Software economics becomes clearer once the unpaid household staff appears in the ledger.
This is also where the phrase “AI-ready” loses some of its conference-hall perfume. A model fed by stale, malformed or carelessly exposed data does not become modern through proximity to artificial intelligence. It becomes a fast consumer of old mistakes. Wilkes’s case is that cleansing, enrichment, governance and verification should happen upstream, as part of a continuous system rather than an apologetic repair crew downstream.
The old contract
Read the transaction log. Apply rows to a target. Restart cleanly after failure. A narrow job, often done well.
The expanded job
Transform, protect, validate, observe and deliver continuously across clouds while preserving context.
04 / Other kinds of timingCode, comedy and the awkward digital twin
Then there is the extracurricular Wilkes. His professional feed occasionally opens a side door onto someone testing the border between engineering and performance. He has discussed releasing an album under the artist name BXLNT. He performs stand-up. In one experiment he combined a script he wrote with an AI-generated version of his face and voice, then evaluated where the synthetic performer handled sarcasm, chuckles and body language well - and where it remained stubbornly wooden.
More recently, he used an AI video tool as a creative partner for brand concepts. He wrote scripts, performed the voiceovers, specified scenes and asked his network to judge the metaphors and storyline. The method sounds less like pressing a magic button than directing a junior studio with infinite patience and occasionally peculiar taste.
These experiments are not evidence that every CTO secretly longs for an open-mic night. They do reveal a consistent appetite: learn a system by making something with it, then expose the result to criticism. Music, comedy and streaming architecture all care about sequence. Deliver a punchline, a note or an alert after its moment and accuracy offers little consolation.
05 / Still movingThe destination keeps changing
Striim’s vocabulary has expanded with the infrastructure around it: multi-cloud movement, schema evolution, continuous validation, AI-ready pipelines. The founding question remains recognizable. Can data be observed and improved while it moves? Can a company shorten the distance between an event and a useful response without sacrificing reliability or control?
Wilkes’s career offers no theatrical conversion from one belief to another. It is a long continuity, from microelectronics to middleware, from replication to cloud integration, from moving records to thinking about them in flight. He followed the transaction between systems and discovered that the space between them was large enough to build a company.
The most human part of the story is its impatience. Yesterday’s answer can be correct, elegant and beautifully formatted. It can also arrive after the customer has left, the machine has failed or the traffic has closed around you. Wilkes has spent much of his working life trying to make “now” a dependable piece of enterprise architecture. The clock, as ever, continues to stream.