Summary

• Atlassian has 46 forward-deployed engineers and says the team is moving towards 100 across the US, EMEA and India.
• The programme has worked with more than 100 enterprise customers and put over 80 AI agents into production.
• Engineers embed with customer teams to identify workflows, build production code and connect AI to existing systems and organisational context.

Atlassian is expanding a team of engineers who work directly inside customer AI projects, making implementation expertise part of the product strategy as enterprises struggle to turn capable models into software that survives production.

The company has 46 forward-deployed engineers and says it is moving towards 100, with the team distributed across the US, EMEA and India. Atlassian reports that the programme has already worked with more than 100 enterprise customers and put more than 80 AI agents into production.

Forward-deployed engineers differ from conventional consultants who diagnose a problem and then leave implementation to someone else. Atlassian describes the role as senior software and applied-AI engineering carried out alongside customers, from identifying useful workflows through writing production code and connecting systems.

The programme began as a small pilot before expanding in the US during June, and the planned increase in headcount indicates that the gap between buying AI capability and using it reliably remains large enough for a major software vendor to allocate scarce engineering talent directly to customers.

AI adoption creates an implementation burden

Enterprise software traditionally becomes more attractive to vendors as delivery becomes repeatable, because one product can serve many customers without equivalent growth in engineering labour. Generative and agentic AI complicate that model when useful deployments depend heavily on organisation-specific data, permissions, workflows and definitions of success.

An agent handling service requests, for example, needs more than access to a capable model. It must know which information sources are authoritative, understand where work is recorded, carry appropriate permissions, recognise cases requiring human review and operate within the customer’s security controls.

Atlassian says its engineers begin with a business outcome and trace the people, systems, knowledge and hand-offs involved before building. Its programme describes discovery, build, adoption and value phases intended to move a use case into production while leaving customer teams able to continue the work.

Company-selected examples indicate how bespoke the process can become. Atlassian says one financial institution moved from uncertainty about a use case to a working agent in a secure sandbox in less than three weeks, while a motorsport organisation progressed from no identified AI use cases to six production agents in fewer than five months.

Those examples do not amount to independent performance evidence, and the programme remains targeted at selected enterprise cloud customers using Atlassian’s AI technology. They do demonstrate why production deployment often resembles a compact systems-integration project rather than simply activating another software feature.

Embedded engineering moves inside the vendor

Techopia recently examined how Devoteam is using embedded engineers to move enterprise AI projects into production. Atlassian’s expansion applies a similar delivery model inside a large software vendor rather than through a separate consultancy.

Because Atlassian owns products including Jira, Confluence and Rovo, recurring customer problems can feed back into product development. A difficult integration encountered repeatedly may eventually become a reusable feature instead of remaining a bespoke piece of engineering work.

That feedback loop creates a tension between tailoring and standardisation. Customers need deployment work specific enough to produce measurable value in their own environment, while Atlassian benefits when the resulting engineering can be generalised across a much larger customer base.

The economics will depend on whether embedded engineers unlock enough software revenue, adoption or reusable product learning to justify a labour-intensive model. Doubling an engineering team can support more complex deployments, but it does not carry the near-zero marginal distribution cost associated with purely software-led expansion.

Atlassian says engagements target measurable value on relatively short timescales and are built around existing permissions, data policies and human oversight rather than unconstrained autonomy. Those controls reflect where many enterprise AI projects now encounter difficulty: not model access, but the work required to integrate AI with the organisation around it.

Model capability continues to improve and access is becoming easier, while companies still have to decide which workflows deserve automation, connect fragmented internal context and satisfy security and operational teams that the resulting system can be trusted.

Atlassian’s move towards 100 forward-deployed engineers therefore exposes an important limit in the current AI software model. Vendors can distribute agents and models rapidly, but turning them into dependable enterprise systems can still require an engineer embedded close to the work.