Our Client
UK-based B2B SaaS specialist IRIS Software Group wanted its teams to embrace Agentic AI capabilities to treble productivity and strengthen its competitiveness. The objective was not cost cutting or even doing the same work faster. It was creating the capacity to take on high-value work that limited bandwidth had previously made impossible.
To enable that higher-value work, IRIS partnered with SoftServe and Anthropic to establish agentic ways of working across Product and other business teams. In a series of four week courses, the programme delivered 60 reusable Claude Skills, 4 plugins, standardised AI workflows, and enabled teams to begin adopting spec-driven development as part of day-to-day delivery.
| Before | During | After |
|---|---|---|
| Strong technology culture | Agentic operating model | Five standard workflows |
| Al tools available | Toolchain standardisation | Shared toolchain |
| No common methodology | Workforce enablement | Repeatable methodology governance |
| Inconsistent adoption | Reusable assets | Behavioural change |
IRIS builds regulation-driven software for accountancy and tax, payroll and HR, education, and bookkeeping automation — markets where accuracy is non-negotiable and customer scrutiny is high. That made it critical to get agentic working right from the outset: adopted as a disciplined engineering capability, not as an experiment.
IRIS already had an embedded technology culture and strong AI adoption signals. But it lacked a shared method, standard toolchain, and structured training to make agentic ways of working consistent, measurable, and scalable across multiple teams.
Hands-on programme
IRIS asked SoftServe to deliver a multi-cohort agentic capability training programme, built around Anthropic’s Claude software, that combined lectures, hands-on workshops, hackathons, and practical assets focused on the full Product Development Lifecycle (PDLC).
But this was not just training content, it also asked SoftServe to help create reusable enablement infrastructure built on Claude. These included the implementation of 60 Claude Skills, 4 plugins, a PM Spec Kit, gate checklists, templates, synthetic practice products, and workflow guidance.
As a result, IRIS replaced ad hoc practices with five repeatable workflows that now underpin agentic delivery. These included:
- Write the spec (to define intent, constraints and success criteria clearly enough for an agent to act on them)
- Shape the task (break work into agent-sized problems)
- Direct the agent (establish commands and repeatable prompts)
- Know when to step in (recognise where automation ends and human judgement on quality and risk begins)
- Review the outcome (model selection, context awareness)
Where it began
From the outset IRIS was clear about its ambitions. "The competitive landscape is tightening and standing still means losing ground. IRIS has resolved to make agentic ways of working the standard across Product and Engineering in FY27."
And the message to its teams was equally clear: "We are aiming to supplement your expertise; not override it. Your engineering and product judgement stays at the centre; agents assist, they do not replace. The aim is to grow output and close ground on competitors."
The key objectives were set out in a pre-project workflow guide to deliver:
- A 3x productivity gain in FY27, measured at team and programme level after a stabilisation window, read alongside defect escape rate and incident counts.
- Agentic working as the standard across Product, Engineering and QA — every team, all divisions, across FY27.
- One standard toolchain: "Claude/Cowork › Aha! › Jira › Cursor › GitHub"
- Reliable measurements, which the toolchain exists to make possible.
How it unfolded
Delivery was global: UK-centred, with significant participation from North American and offshore teams. The programme itself was delivered to each cohort remotely across time zones.
Each cohort received different practice products that were deliberately matched to the division being trained. and shipped with its own full evidence pack, including product idea brief, assumptions register, interview transcripts, meeting notes and messy raw notes, allowing participants to practise on realistic mess rather than clean inputs. They were divided into product and cross-stream sessions.
In the product stream each course unfolded over four weeks.
For the cross-stream elements every session was pre-booked with a stated agenda and takeaways, slides shared upfront and recordings centralised rather than held in individual accounts. The tool strategy explicitly stated: "Claude Chat/Cowork/Design, Jira/ADO, Aha! and connectors integrate and where there is and is not duplication".
The technologies used included:
Claude Chat/Cowork/Design
Agentic assistant used by Product for knowledge work, spec drafting and orchestration across the lifecycle.
Aha!
To be “Single Source of Truth” in the future -"Product roadmap, prioritization, initiatives, epics and features. Source of truth for product strategy and GTM process."
Jira
Engineering execution layer: sprints and workflow tracking. Linked to Aha! for idea-to-delivery traceability.
Outcomes
The project succeeded because, in IRIS’s own words, SoftServe coached rather than lectured. It also built tooling, not just content, as reflected in the 60 packaged Claude Skills, 4 plugins, and a documented brownfield codebase. It meant participants not only left with working equipment, but with knowledge depth on the platform, while IRIS had a formal Anthropic partnership and Claude-knowledgeable architects.
The results showed up quickly. The programme reached 100% adoption of the agentic toolchain among trained Product participants, making Claude-driven work the standard rather than the exception. Product teams now use Claude to generate requirements 2.5x faster from initiation to review, and one engineering team attributed a significant reduction in delivery time on a complex backend decoupling task directly to the AI-assisted workflows introduced during training. Across the PDLC, throughput rose 22%, with the best-performing teams reaching 2x velocity. Delivery satisfaction was high, with around 85% of Cohort 2 participants saying they would recommend the programme to a colleague.
By the conclusion, IRIS had moved beyond experimentation and established the foundations for enterprise-scale agentic delivery. With standardised workflows, reusable AI assets, integrated tooling, and growing adoption across all product and other teams, the organisation was now positioned to measure and realise productivity gains across FY27.
"SoftServe helped IRIS establish an enterprise-scale agentic operating model that changed how Product, Engineering and QA teams work."
All was completed in a series of four-week, multi-cohort agentic capability training programmes delivered by the SoftServe team in partnership with Anthropic and is poised to create the 3x productivity improvements that IRIS has targeted for FY27.
