How AI is accelerating Salesforce development in WSOY’s environment
Over a year of AI-assisted Salesforce development: what changed and what didn’t?
Ceili has been building WSOY’s Salesforce environment with the help of AI for more than a year. The project has become a living example of what generative AI actually changes in software development, and what it doesn’t.
The pace has clearly picked up. The team delivers more functionality in less time, and AI writes most of the code. Even so, the project isn’t finishing on the kind of timelines that the boldest predictions for AI-driven projects have promised.
The rest of the world has noticed the same thing. Researchers at MIT and Wharton used data from more than 100,000 GitHub developers to study how different generations of AI affect productivity (Writing Code vs. Shipping Code, NBER 2026). They found that autonomous AI coding agents increased coding activity by 180%, but the gain shrank quickly further down the delivery chain. In software that was actually released, growth was only 30%. The headline puts the tension well: writing code and shipping code are two different things.
Why does this happen?
The Novelty Bottleneck framework, published in March 2026, explains it using Amdahl’s law. When AI handles the repetitive parts, the bottleneck becomes the part it can’t replace: human judgement, understanding requirements, and designing the overall architecture. “AI makes you faster, but a task twice as big still means roughly twice the human work.” Better agents reduce how much human work is needed, but they don’t change how that work scales.
Enterprise systems are a particularly good example. Businesses run on them: they price customers, create quotes and manage contracts. A system like this has to be reliable, traceable and accountable, not just functional. You don’t get those qualities by writing code faster. You get them by building things right.
AI and quality in practice: the WSOY case
Werner Söderström Osakeyhtiö is one of Finland’s largest publishers. Its B2B sales team sells books to bookstores and retailers, and Salesforce is at the heart of their sales. It holds quotes, customer accounts and integrations, among other things.
During the project we built two features that made a real difference to the sales team’s daily work:
- A new quoting tool lets salespeople browse the full product catalogue directly in Salesforce, filter products as needed and build quotes quickly, without jumping between several systems.
- Customer-specific pricing puts contract-based discounts and campaign pricing directly in salespeople’s hands, removing manual work and separate spreadsheets.
Even when most of the code is generated with AI, the core of the project is still very human-centred. AI often helps design features, but people set the direction and make the final decisions. Code produced by an agent never goes to production on the first attempt. A developer reviews it first to make sure it follows the agreed practices and meets the business’s real needs.
We guide the AI with instructions and examples attached to the project, so the code it produces fits the rest of the implementation from the start. This is one of our key lessons: the more clearly a developer can describe the goal and the constraints, the more usable the first version is.
A few highlights from the project, and what they mean in practice
"AI has written all the code since the start of the year"
This doesn’t mean the developer has left the picture. The developer’s role has shifted from producing code to reviewing and directing it. Before implementation starts, we make sure the requirements are clear: what the user needs, how data moves between systems, and what the business-critical constraints are. Once that’s settled, AI produces a first version quickly, and the developer can spend their time where professional expertise really matters.
Before a change is approved, the code always goes through review: at least one other team member checks that the implementation meets the business’s needs. AI doesn’t remove this responsibility. It frees up time to do it more carefully.
"Quality has stayed the same or even improved since we started using AI"
Quality doesn’t improve on its own. It improves because the team keeps the same quality gates as before. Repetitive work now gets done faster, which leaves more time for those gates. Speed hasn’t replaced care. It has given care more room.
"The current implementation would have taken significantly longer without AI in development"
Sprint throughput has improved by about 30%, so more gets finished in the planned time.
What have we learned?
A developer’s day-to-day work has changed a lot. The role has moved from producing code to reviewing it and shaping the whole solution. This doesn’t reduce the need for professionals. It changes where their expertise goes.
A large share of the work still happens away from the computer. Understanding business requirements, working with WSOY’s experts, agreeing integrations with other systems and training users all move at their own pace, through people talking to each other. Faster coding doesn’t automatically shorten a project, not if it would mean cutting corners where careful thinking matters most.
The world has moved, or is moving, from maximising efficiency to improving quality
In the past, time pressure might have pushed us towards a technically lighter solution. Now we can be more ambitious because we don’t run out of time. The result serves WSOY’s salespeople better. It isn’t just done, it’s done right.
The further we’ve come, the more clearly we’ve seen a shift in professional identity. AI hasn’t made developers redundant. It has strengthened their role as people who see the big picture and build lasting solutions. Working with AI has taught us to direct and evaluate its output more and more critically. The real value of expertise is the ability to build solutions that are secure, scalable and serve their users reliably over the long term. AI can’t do that alone, but with it you can get a lot further.
Meet WSOY’s Verkkokauppa Verneri agent!
Ceili is building an AI agent to streamline onboarding for WSOY’s new B2B customers. We started by testing the idea on a smaller step that could be validated quickly: the Verkkokauppa Verneri (“Webshop Verneri”) agent, which suggests suitable products for different B2B campaigns.
Author:
Otto Heikinmäki
Developer Team Lead &
Senior Salesforce Developer
Read more
Marketing Cloud Next implementation: how the project unfolds with Ceili
Ceili has delivered numerous Salesforce Marketing Cloud Next and Data 360 implementations for organizations across the board: from manufacturing and professional services to IT, financial
Ceili Winter ’27 Release Notes
Intro Here is Ceili’s traditional summary of the upcoming new features and changes in Salesforce’s Winter ’27 release. Salesforce publishes an update three times a
AI-Assisted Salesforce Development at Ceili
AI-assisted Salesforce development makes the design, build, testing and documentation of Salesforce projects more efficient. At Ceili, AI supports our experts in Salesforce implementations, ongoing
Salesforce®, Sales Cloud® ja muut ovat salesforce.com, inc:n tavaramerkkejä, joita käytetään täällä luvan kanssa.
