Contract specialists can move AI work forward quickly. When they also own work the company depends on every day, the company loses that knowledge as soon as the contract ends.
Only 19% of companies have changed their mix of full-time, contract and gig workers in response to AI, according to Deloitte’s State of AI in the Enterprise 2026. Just 36% are hiring specialized talent to drive AI initiatives. Most companies are staffing AI work the same way they staffed everything else, either waiting for full-time hires or bringing in contractors case by case. Neither approach answers the question that matters: which parts of the work need to stay inside the company after launch. Contracts work well for time-bound builds. The people who own the model, the data and the outcome should be full-time employees.
Key findings
Only 19% of companies have changed their mix of full-time, contract and gig workers in response to AI, per Deloitte.
36% of companies are hiring specialized talent to drive AI initiatives, while 53% rely mainly on training their existing workforce.
AI tools bought from vendors or built with partners succeeded about 67% of the time in MIT’s research, compared with about a third as often for internal builds.
68% of large US companies say limits on hiring and upskilling are shaping their AI strategy, per KPMG.
More than three-quarters of tech executives have seen employees finish training and still be unable to apply the skills on the job, per Pluralsight.
Most companies haven’t decided
Deloitte’s data shows the gap. Companies name the skills gap as their biggest barrier to AI, yet under a quarter have rethought the balance of full-time and contract work. The rest make staffing decisions one project at a time, usually based on what’s fastest to approve. That’s often a contractor when headcount is frozen and a full-time requisition when it isn’t.
The pressure is real. KPMG’s Q2 2026 AI Pulse Survey found that 68% of large US companies say limits on hiring and upskilling are influencing their AI strategy. Training alone won’t close that gap on a project deadline. Pluralsight’s 2026 Tech Skills Report found that more than three-quarters of tech executives have seen employees finish training or earn certifications and still struggle to apply the skills on the job.
So companies need outside expertise, and the evidence suggests it works. In MIT’s NANDA research, AI tools bought from vendors or built with partners succeeded about 67% of the time, while internal builds succeeded about a third as often. Experienced outsiders bring production experience that internal pilot teams usually lack. The open question is what happens to that experience after the contract ends.
When contract is the right choice
Contract specialists fit work with a clear end point and a defined deliverable. Common examples in AI work include:
Integrations and builds. Connecting to a new buying protocol like the ones covered in our look at agentic ad buying , wiring up a clean room, or building a data onboarding pipeline.
The push from pilot to production. MLOps setup, evaluation frameworks and monitoring are often the work that stalls AI pilots , and specialists who have done it before can set it up quickly.
Specialized one-off problems. A model audit, a performance optimization or a migration that needs deep expertise for a few months, not years.
Covering a gap while a search runs. A technical hire takes about 10 weeks to fill, and a contractor can keep critical-path work moving in the meantime.
These tasks share a pattern. The skill is scarce, the timeline is short, and once the work is finished, someone else can run it.
When full-time is the right choice
Some AI work shouldn’t be contracted out, no matter how hard the hire is.
The people who own the model. They decide what it optimizes for, when it gets retrained, and when to override it.
The people who own the data. In ad tech, that means identity, consent and clean-room pipelines, which hold the institutional knowledge the rest of the AI stack depends on.
The business owner accountable for outcomes. Daily accountability for what an agent or model does belongs to someone who stays.
Anyone on call for a production system. When a bidding model starts overspending on a Friday night, the person who answers can’t be one whose contract ended last month.
These roles build knowledge that adds up over time. They’re also how a company develops its own senior talent. Leaders have warned that AI can erode the hands-on work that turns junior engineers into experts. If the core of the work goes to contractors, the company loses the ability to develop those experts in-house.
The failure mode is contracting the core
The most expensive mistake is letting a contractor become the owner by default. It usually starts reasonably. A specialist is brought in for a three-month build, the build works, the system goes live, and the specialist stays on to maintain it because nobody else understands it. A year later, the company’s most important model or pipeline is maintained by someone with no long-term stake in it and no one inside who could take over.
Contracting the core also carries legal risk. The US Department of Labor’s proposed rule on contractor classification keeps the permanence of the working relationship and whether the work is central to the business among the factors that decide whether a contractor should be treated as an employee. Many states apply their own, stricter tests. A contractor who has run a core system for years with no end date in sight starts to look like an employee under several of those factors.
The opposite mistake costs time. Hiring a full-time engineer for a four-month integration means waiting about 10 weeks for someone who may not have enough work once the integration is done.
Pair every contractor with an owner
Teams that use contract specialists well build knowledge transfer into the engagement from the start.
Name a full-time owner before the contract begins. That person co-builds with the contractor and takes over when the contract ends.
Write knowledge transfer into the statement of work. Make documentation, runbooks, recorded walkthroughs and a handover period deliverables, not extras.
Overlap at the end. Plan for a few weeks where the owner runs the system and the contractor supports.
Set an end date and enforce it. If the work turns out to be ongoing, convert it into a full-time role or move it to a full-time owner rather than renewing the contract indefinitely.
Track the mix. Treat contractors as part of the four-segment workforce most companies now run, with each engagement visible to the business owner and to HR.
Contract versus full-time is the wrong way to frame the decision. The better question is what the company needs to keep. Contract the build and keep the knowledge.
Frequently asked questions
Should we hire AI engineers on contract or full-time? Full-time for people who own models, data, production systems and business outcomes. Contract for work with a clear end date, such as integrations, the push from pilot to production, or a specialized one-off problem.
What AI work is best suited to contractors? Integrations, clean room and data onboarding builds, MLOps and evaluation setup, model audits, migrations, and critical-path work while a full-time search is under way.
What AI roles should be full-time? Model owners, data and identity owners, the business owner accountable for outcomes, and anyone on call for a production system.
Are AI projects built with outsourced partners more successful? In MIT’s research, AI tools bought from vendors or built with partners succeeded about 67% of the time, compared with about a third as often for internal builds. Experienced partners bring production skills internal teams often lack.
How do we keep knowledge when an AI contractor leaves? Pair each contractor with a named full-time owner, write documentation and a handover period into the statement of work, and plan for overlap before the contract ends.
Can long-term contractors create legal risk? Yes. US federal guidance and many state laws consider the permanence of the relationship and whether the work is central to the business when deciding whether a contractor should be treated as an employee.
Is it faster to bring in an AI contractor than to hire full-time? Usually. A full-time technical hire takes about 10 weeks, and a contract specialist can often start critical-path work while that search runs.
How many companies are changing their mix of contract and full-time workers because of AI? Only 19%, according to Deloitte’s 2026 State of AI in the Enterprise survey.
When should a contract role become full-time? When the work turns out to be ongoing, when the person has become the only one who understands a system, or when the role starts to carry accountability for business outcomes.
