All blog posts
CI failure on mainGitHubJenkinsSlack
Delivers: A pull request with the fix and 50 green runs in a row
Software engineering
Flaky Test Fixer
Reproduces each flaky test, finds the race and fixes it.
Labelbox is partnering with Wipro across two products: Horizon, reinforcement learning environments and evaluations for frontier models, and Recursion, managed enterprise agents. One helps train the models. The other puts agents to work inside the enterprise, on work that carries real responsibility: investigating production failures, reviewing code for security issues, identifying problems in datasets, and running experiments that lead to actionable recommendations.
Wipro brings deep enterprise relationships and experience running mission-critical systems across industries, with more than 230,000 employees and business partners in 65 countries. Labelbox brings the technology to execute complex workflows, measure outcomes, and continuously improve AI systems through experience.
Labelbox’s reinforcement learning capabilities will integrate with Wipro Intelligence™, Wipro’s unified suite of AI-powered platforms, solutions, and transformative offerings. Wipro Ventures, Wipro’s corporate investment arm, also participated in Labelbox’s recent Series E.
Horizon· WorldSim
Horizon· WorldSimSimulated worlds with real work
WorldSim builds the company your model will work in.
Code, dashboards, spreadsheets, decks and docs, filled with realistic business data. Each scenario is its own world.
Inside one world, the model works a hard task end to end.
One task crosses a dashboard, logs, code, a spreadsheet, slides and a doc. Every step is recorded as a trajectory.
Verifiers decide what’s correct. Experts decide what’s good.
Deterministic checks score every trajectory the same way, every run. When two are both correct, an expert’s preference teaches the model taste.
Thousands of worlds. Millions of graded trajectories.
Coding, computer use, research, multimodal and voice, tuned to your pass@k targets.
The model learns the work, not the shortcut.
Verifiable reward for correctness and expert preference for taste, at the scale post-training needs.
Wipro’s global reach, Labelbox’s learning systems
Powerful models are now widely available. But models alone don’t understand how your business operates, what your experts know, or what defines a good outcome. The advantage belongs to companies that combine models with their own knowledge, workflows, systems, and standards. And that advantage compounds as AI systems learn from reviewed work and improve with every cycle.
Getting there takes more than deploying a model. It requires identifying the right workflows, connecting systems, capturing expert knowledge, and continuously measuring performance.
That’s what this partnership is built to do. Wipro brings its global reach, with engineers close to customers wherever they operate. Labelbox brings Horizon, where models train on environments that mirror real work, and Recursion, where agents learn from every run. Together, we can take AI that improves with experience to enterprises at scale.
Horizon: Reinforcement learning environments for frontier models
Models learning to do complex work need places to practice. Labelbox Horizon provides reinforcement learning environments and evaluations for reasoning, coding, tool use, and computer use. Its WorldSim technology recreates enterprise applications and realistic business scenarios, with verification and expert judgment supplying the training signal.
Wipro plans to use Horizon in its work with hyperscalers and frontier AI labs. The goal is to shorten the path from a new research requirement to a program ready to run, with environments built, tasks defined, and evaluation criteria in place.
Recursion: Managed enterprise agents that work to your standards
Teams define a goal, the tools an agent can use, and the standards its work must meet. Agents can start on a schedule or in response to an event, coordinate specialists, and work across business applications. Every result comes with supporting evidence and an evaluation against the team’s criteria.
Take a goal that takes real engineering judgment: cut a ranking model’s latency by 30% without losing accuracy. A coordinator agent splits it into experiments and hands each to a specialist, some of which wait on GPUs or on each other. A separate grader then scores the recommendation against the team’s own criteria, with the evidence attached.
Recursion Managed Agents· Sessions
Recursion Managed Agents· Sessionscoordinator · 6 specialists · 1 grader
Inference coordinatorClaude
JiraENG-2291
Cut the ranking model’s p95 latency by 30% without losing accuracy.
planreading ENG-2291 and the code
Profiler
Quantization
Batching
Distillation
Load test
Accuracy guard
recommendation
Ship int8 with a 4 ms batch window: p95 −38%, accuracy −0.1.
ProfilerGPTDatadog
not started
QuantizationClaudeGitHubSandbox
not started
BatchingOpen weightsGitHubSandbox
not started
DistillationGeminiGoogle Cloud
not started
Load testGPTSandboxGrafana
not started
Accuracy guardClaudeGoogle BigQuerySandbox
not started
Outcome graderClaude
your rubricwaiting for a recommendation
Bottleneck found in production
evidence: Datadog traces, 10k requests
Change is tested and reviewed
evidence: PR #5120, CI green
No new hardware
evidence: same GPU pool, 2.3× throughput
p95 down at least 30%
evidence: load test at 2× peak traffic
Accuracy within 0.2 points
evidence: 50k held-out queries, 3 seeds
satisfied 5/5ship to a 5% canary
An example Recursion session: one latency goal, six specialists, graded against five criteria.
Agents keep lessons that have been reviewed, so they get better at the workflows they repeat. People set the standards and control permissions and sensitive actions.
Wipro’s forward-deployed engineers will help customers pick the right workflows, connect their systems, capture domain expertise, and define how performance is measured.
Recursion is built for the most demanding recurring work across the enterprise: work that needs real expertise and that teams never have enough people to keep up with. Agents work around the clock, many sessions in parallel, and hand back finished work graded against your standards.
Here are some of the agents we run ourselves at Labelbox, from incident response and security review to experiment sweeps and research.
What’s next
Together, Horizon and Recursion help AI systems get tested, governed, and improved through experience, whether the system is a frontier model in training or an agent doing a team’s work.
“As enterprises move towards more advanced AI systems, reinforcement learning will play an increasingly important role in developing models capable of addressing complex, real-world challenges. By bringing Labelbox’s reinforcement learning capabilities with Wipro Intelligence™, and pairing them with our FDEs, we are giving clients a faster, governed path from promising pilots to production results they can measure.”
Every company has expertise that took decades to build: how to diagnose an unusual failure, which signals matter in an experiment, when to escalate. Wipro and Labelbox are bringing that expertise into AI systems that learn and improve with experience.
Want to explore what’s possible for your organization? Get in touch with us to shape the next generation of intelligent enterprise systems.