--- title: How to Build an AI Workforce of People and Agents description: "How to build an AI workforce: which work AI agents should own, where people stay essential, and the roles, controls and team you need around the agents." image: https://blog.potentiam.co.uk/hubfs/blog/images/how-to-build-an-ai-workforce-hero-20260930-165319.webp --- [![footer-logo](https://blog.potentiam.co.uk/hubfs/footer-logo.svg "footer-logo")](https://potentiam.co.uk) - [Solutions](https://www.potentiam.co.uk/solutions) [![Technology icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/Technology%20icon.svg) Technology Scale your team with a dedicated advisor who aligns architecture and performance 
to your goals.](https://www.potentiam.co.uk/solutions/technology) [![Finance and Accounting icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/Finance%20and%20Accounting%20icon.svg) Finance and Accounting Finance specialists build offshore teams using the right processes and skills to strengthen accounting and deliver results.](https://www.potentiam.co.uk/solutions/finance-accounting) [![Marketing and Creative](https://blog.potentiam.co.uk/hubfs/Images/Icons/Marketing%20and%20Creative.svg) Marketing & Creative Marketing advisors build offshore teams that close capability gaps and integrate seamlessly with local operations.](https://www.potentiam.co.uk/solutions/marketing-and-design) [![Data Analytics icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/Data%20Analytics%20icon.svg) Data Analytics Analytics experts design scalable offshore data teams that deliver actionable insights and business value.](https://www.potentiam.co.uk/solutions/data-analytics) [![Sales and Customer Success icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/Sales%20and%20Customer%20Success%20icon.svg) Sales & Customer Success Specialists build offshore sales and customer success teams aligned for growth and strong client relationships.](https://www.potentiam.co.uk/solutions/sales-customer-service-solutions) [![Engineering and Construction icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/Engineering%20and%20Construction%20icon.svg) Energy Consulting Solutions Scale engineering and construction projects with offshore teams that improve control, efficiency, and on-time delivery.](https://www.potentiam.co.uk/solutions/energy-consulting) - Locations [![menu-icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/menu-icon.svg) Cape Town, South Africa English-speaking teams with seamless UK collaboration](https://www.potentiam.co.uk/locations/south-africa) [![menu-icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/menu-icon.svg) Bengaluru, India Technology and data capability with EU and UK time-zone overlap](https://www.potentiam.co.uk/locations/india) [![menu-icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/menu-icon.svg) Iași, Romania Multilingual EU talent aligned with European markets](https://www.potentiam.co.uk/locations/romania) [![menu-icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/menu-icon.svg) Johannesburg, South Africa English-speaking talent with UK time-zone alignment](https://www.potentiam.co.uk/locations/johannesburg) [![menu-icon](https://blog.potentiam.co.uk/hubfs/Images/Icons/menu-icon.svg) São Paulo, Brazil Latin American talent with US and EU time-zone overlap](https://www.potentiam.co.uk/locations/brazil) - [Careers](https://www.potentiam.co.uk/careers) - [About](https://www.potentiam.co.uk/about-us) - Resources [![Blog](https://blog.potentiam.co.uk/hubfs/Blog.svg) Blogs Explore strategies and insights on building high-performing offshore teams, scaling global operations, and unlocking performance growth.](https://blog.potentiam.co.uk) [![case guides](https://blog.potentiam.co.uk/hubfs/case%20guides.svg) Guides Explore essential offshoring strategies for IT managed service providers with our comprehensive guides tailored for COOs and CEOs.](https://www.potentiam.co.uk/guides) [![webinar](https://blog.potentiam.co.uk/hubfs/webinar.svg) Webinars Watch Potentiam’s webinars to discover how offshore teams can accelerate growth, cut costs, and boost performance.](https://www.potentiam.co.uk/webinars) [![case studies](https://blog.potentiam.co.uk/hubfs/case%20studies.svg) Case Studies Explore case studies showcasing how companies in various industries are enhancing efficiency and growth through innovative strategies and specialised services.](https://casestudies.potentiam.co.uk) - [Contact](https://www.potentiam.co.uk/contact-us) - [Talk to Us](https://www.potentiam.co.uk/contact-us) Back to Articles ![Operations director and finance lead in a London office sorting process cards on a table as they decide which workflow steps AI agents will handle and which stay with people](https://blog.potentiam.co.uk/hubfs/blog/images/how-to-build-an-ai-workforce-hero-20260930-165319.webp) Scroll Back to Articles [IT & Software](https://blog.potentiam.co.uk/tag/it-software) | 09 November 2026 # How to Build an AI Workforce of People and Agents [Potentiam](https://blog.potentiam.co.uk/author/potentiam) An AI workforce is a team of people and AI agents organised around the same work. The agents carry out defined, repeatable, multi-step tasks inside your systems. People own the judgement, the relationships, the exceptions and the results. Building one is mainly an operating model decision: in our experience, the businesses that get value from agents decide, workflow by workflow, what each agent owns, who owns each agent and who handles what it cannot. Our advice is to start with one workflow that is high-volume, rules-based and measurable, give the agent a named human owner and a clear escalation route, and staff the work around it as a permanent team: the data, the testing, the monitoring and the exceptions. Our first article in this series covered [how AI is changing what companies offshore](https://blog.potentiam.co.uk/ai-changing-what-companies-offshore); this one is about designing the workforce of people and agents. Key takeaway The agentic layer is the set of AI agents doing defined work across your processes, plus the data access, limits and people that let them do it safely. Give agents work that is repetitive, rules-based, high-volume, system-driven and measurable. Keep judgement, relationships, consequential decisions and accountability with people. Every agent needs a human owner, a documented job, set permissions and an escalation route. Capacity gains depend on staffing that human work. AI changes the work, and that work still needs an effective team around it. ## From AI tools to an AI workforce For most businesses the first wave of generative AI was personal: assistants that draft, summarise and answer questions for one person at a time. An AI agent, given a goal and access to your systems, can read an incoming request, look up and update the records, draft a reply and pass the case on. That makes AI capacity that works for the organisation, and it has to be designed, supervised and improved like the rest of the workforce. Larger organisations are moving first. In [McKinsey's State of AI 2026](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai) (August 2026, 1,719 respondents worldwide), 40% of respondents from organisations with more than $1 billion in annual revenue reported scaling AI agents in at least one function, up from 27% a year earlier; among smaller organisations the share was flat at 22%. [Gartner's September 2026 forecast](https://www.gartner.com/en/newsroom/press-releases/2026-09-16-gartner-forecasts-worldwide-ai-spending-to-grow-49-point-5-percent-in-2026) expects worldwide spending on AI agents and assistants to rise from about $16.5 billion in 2025 to $29.2 billion in 2026 and $65.5 billion in 2027, as software vendors embed agentic AI in their existing products. For a mid-sized UK business, that means many agents will arrive inside the finance, CRM and service software you already use, planned or not. Most smaller firms are still at the start. In the [OECD's 2026 survey of more than 2,000 SMEs](https://www.oecd.org/content/dam/oecd/en/publications/reports/2026/04/empowering-smes-in-the-age-of-ai_7f58652c/bf5a9816-en.pdf) in 12 countries (April 2026, a non-representative sample), 75% used off-the-shelf AI applications and 3.6% had deployed agentic AI. That leaves time to settle the workforce design before the agents multiply. ## What is the agentic layer? The agentic layer is the set of AI agents working across your business processes, together with what they need to run: access to data and systems, the rules and limits they work within, and the people who own, check and improve them. It has three parts. The **agents** do defined work. The **controls** set what each agent may see and do, what it must escalate and what gets logged. The **human layer** owns each agent, handles what it passes on, checks its output and fixes the workflow when it goes wrong. Unlike the scripted bots of an earlier digital workforce, agents cope with variation, so they are more useful, less predictable and more dependent on the human layer. ## Which work should AI agents own? Give an agent a workflow when it passes five tests: 1. **Repetitive:** the same kind of task comes round many times a day or week. 2. **Rules-based:** you can write down what a correct outcome looks like. 3. **High-volume:** there is enough of it to repay the set-up and supervision. 4. **System-driven:** the inputs and outputs live in systems the agent can reach with the right permissions. 5. **Measurable:** you can check the result against a standard and count the errors. Typical candidates are matching invoices to orders, updating CRM records and routing service tickets. Treat the five tests as a first screen. Where all five hold, the agent can own the routine path once it has proved itself in testing, with its permissions approved and human sign-off kept for high-impact or irreversible actions, and it hands anything that fails the rules to a person. Where a workflow passes only two or three, use the agent to prepare the work and a person to finish it. Staff are more open to this than the headlines suggest. In a [Stanford study of 1,500 US workers](https://arxiv.org/abs/2506.06576) in 104 occupations (June 2025, preprint), workers wanted AI agents to automate 46.1% of their tasks, yet in 47 occupations the preferred arrangement was an equal partnership between worker and agent. ## Where people remain essential ![Customer operations specialist in a Cape Town office reviewing a printed case file with a senior colleague to resolve an exception an automated process could not handle](https://blog.potentiam.co.uk/hubfs/blog/images/how-to-build-an-ai-workforce-exception-handling-20260930-165330.webp) In the workflows we see, five kinds of work stay with people. **Judgement**, where the right answer depends on context the rules do not capture. **Relationships** where trust is at stake. **Complex decisions** with financial, legal or customer consequences. **Accountability**, which needs a named person. And **oversight**: checking what the agents do and deciding when to change it. Agents are improving fast and still make mistakes: the [Stanford AI Index 2026](https://hai.stanford.edu/ai-index/2026-ai-index-report) (April 2026) reports agents' success on OSWorld, a benchmark of real computer tasks, rising from about 12% to about 66% in a year. In a [September 2026 analysis](https://www.gartner.com/en/articles/agentic-ai-infrastructure-governance), Gartner recommends giving every agent a responsible human owner, letting low-risk activity run autonomously and requiring explicit human approval for high-impact and irreversible actions. It predicts that by 2029 at least 70% of organisations running agentic AI in production in IT infrastructure and operations will have a material service, security or cost incident linked partly to insufficient runtime controls. How people step in matters as much as whether they do. In a [randomised field experiment in customer service on Alibaba's Taobao marketplace](https://arxiv.org/abs/2605.14830) (working paper, May 2026), an agentic AI system cut average chat duration but substantially lowered customer ratings on the chats it handled. Human intervention preserved service quality when the agent hit a problem beyond its capability, worked less well once customers were frustrated, and depended on stepping in early. Design escalation to bring people in early, and size review work like any other workload. ## Building the operating model around the agents Before an agent goes live, write it a one-page job description: its purpose, the systems and data it may read or change, its limits (value thresholds, customer types, hours), what it escalates and to whom, how it is measured and who owns it. If you cannot fill in the page, the workflow is not ready for an agent. Then settle five things across the agentic layer as a whole: - **Data:** which sources each agent relies on, who owns them and who fixes them when they are wrong. - **Systems and permissions:** the minimum access each agent needs, granted and reviewed like a new starter's. - **Governance:** the limits on autonomy for each workflow, and who can change them. - **Monitoring:** accuracy, volumes, cost and incidents, reviewed on a set rhythm by a named person. - **Escalation:** a staffed queue for what agents pass on, with response times that suit your customers. Microsoft's [2026 Work Trend Index](https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization) (May 2026; vendor research surveying 20,000 AI users in ten countries including the UK) poses three questions: who reviews agent performance, who can update the workflows agents run, and how a local improvement gets shared. We would add a fourth: who picks up the escalation queue at eight in the evening. ## Building the human capability around AI The people around an agent workforce do recognisable jobs. This table follows an agent-run workflow from design to improvement, with the roles we build for each stage. | Stage | What people do | Roles that typically do it | | --- | --- | --- | | **Design** | Map the workflow, decide what the agent owns and write its job description, limits and escalation rules | Business analysts, with your process owner | | **Build and connect** | Configure the agent, connect it to your systems with the right permissions and maintain those connections | AI and machine learning engineers, data engineers | | **Data** | Keep the records the agent reads and writes accurate, complete and consistent across systems | Data quality specialists, data analysts | | **Testing** | Test the workflow before launch and after every change, and sample live output against an agreed standard | QA engineers | | **Run and monitor** | Watch accuracy, volumes, cost and failures, handle incidents and apply vendor updates | Site reliability engineers, application support analysts, BI developers for the reporting | | **Exceptions** | Resolve what the agent escalates and talk to the customer or supplier where needed | Customer operations specialists, and the finance or service staff who own the process | | **Security** | Review agent access, watch for misuse and respond to security incidents | SOC analysts, security engineers | | **Improve** | Find out why exceptions happen, remove the causes and widen what the agent handles once it has earned it | Business analysts, data analysts | We can build all of these roles in any of our hubs. Which hub suits a team depends on the work, the hours it needs to cover and the skills involved, and we agree that with each business. Demand is shifting towards judgement. In the same Work Trend Index survey, AI users named quality control of AI output (50%) and critical thinking (46%) as the human skills that matter more as AI does more. [PwC's 2026 Global AI Jobs Barometer](https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-jobs-barometer.html) (June 2026) found, in 2.4 million US entry-level job adverts, that the entry-level roles most exposed to AI were seven times more likely than the least exposed to require traditionally senior skills such as leadership and creativity. Plan for more people who check, decide and improve. ## Unlocking workforce capacity ![QA engineer, data analyst, business analyst and team lead in an Iași office holding a morning stand-up to agree fixes to an automated workflow](https://blog.potentiam.co.uk/hubfs/blog/images/how-to-build-an-ai-workforce-team-review-20260930-165340.webp) Agents can increase the amount of work a team handles, and the gain is set by the slowest human stage. A [study of more than 500,000 software developers](https://www.nber.org/papers/w35275) (NBER working paper, May 2026, revised September 2026) found that autonomous coding agents increased commits (recorded code changes) by 240%, but the effect fell to 80% for the number of projects and to 30% for the number of releases. The authors attribute the drop to human bottlenecks further down the chain. The lesson carries beyond software: if agents triple the volume reaching your reviewers and exception handlers, output rises only as far as they can absorb it. As an illustration: a customer operations team of eight handles 3,000 enquiries a month. An agent takes over routine requests such as order status and address changes and resolves two-thirds of all enquiries end to end. The team now handles about 1,000, all of them the harder cases. Its workload does not fall by two-thirds, because harder cases take longer and someone has to check the agent's replies and maintain what it relies on. Measure handling, review and exception work for a few months before counting spare capacity. If there is some, the business can take on more customers without hiring at the same rate, answer faster, or move people into retention work. Headcount has so far moved less than many expected, although the evidence covers AI in general rather than agents specifically. In McKinsey's 2026 survey, 14% of respondents from organisations using AI said it had contributed to a fall in total headcount over the past year, and two-thirds reported little or no change. In our view the bigger change is in the mix of work, and we advise planning for it: fewer people in repetitive processing, more in exceptions, quality, data and improvement, and managers who supervise people and agents. Decide in advance whether freed capacity goes to growth, service or cost. ## How an embedded team can support the model The work around an agent workforce is continuing and specialist, and in our experience a specialist technical role takes six to twelve weeks or longer to fill in the UK. We build embedded teams in five hubs in four countries: Cape Town and Johannesburg in South Africa, Iași in Romania, Bengaluru in India and São Paulo in Brazil, typically hiring in four to six weeks (four to eight in São Paulo) from starting the search to the candidate accepting the offer. Under [the Embedded Offshore Team Model](https://blog.potentiam.co.uk/embedded-offshore-team-model), you stay in control of the work. Your managers direct priorities, workflows and technical delivery, while we provide the local management and infrastructure that support the team day to day. The people who check the agents, fix the data and clear the exceptions work in your systems, to your standards, so what they learn about how your workflows fail stays with your business. Much of this is work we already build teams for. When [Evora Global's](https://blog.potentiam.co.uk/how-evora-global-saved-12-fte-with-multi-hub-offshore-teams-romania-india) consultants were spending too much of their time on data work, we built a data and operations team in Iași and a software development team in Bengaluru; the Iași team took weeks off the effort of getting data to Evora's clients. For [Opinium](https://blog.potentiam.co.uk/how-opinium-research-built-offshore-data-analytics-team-cape-town), we built a Cape Town team for data processing and cleaning, which over a 12-month period returned about 20 days a month to insight consultants in London, New York and Amsterdam. When [Open Energy Market (OEM)](https://blog.potentiam.co.uk/how-open-energy-market-accelerated-development-offshore-team-bangalore) needed to build its software before revenue caught up, we built it a dedicated development team in Bengaluru. Agents depend on exactly this kind of work. Agents keep working after your UK team goes home, and time zones can help. Bengaluru is four and a half hours ahead of the UK in summer and five and a half in winter, so a 9 to 6 day there runs from 4.30am to 1.30pm UK time in summer and 3.30am to 12.30pm in winter: with the scope agreed, exceptions an agent raised overnight can be cleared before your UK team starts. Full-day or out-of-hours cover needs agreed shifts in any hub. Read more on [the best offshore location for a technology team](https://blog.potentiam.co.uk/best-offshore-location-technology-msp-team) and [why most AI implementation fails](https://blog.potentiam.co.uk/why-most-ai-implementation-fails-potentiam). If an offshore team will see personal data, plan for it at the start. Giving a team employed by a separate legal entity access to UK personal data is usually a [restricted transfer](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/international-transfers/a-brief-guide-to-international-transfers/) under UK GDPR. Romania is covered by UK adequacy regulations, though your other UK GDPR duties still apply; South Africa, India and Brazil need a safeguard such as the International Data Transfer Agreement plus a transfer risk assessment. Take advice on your own position. [Book a discovery call](https://www.potentiam.co.uk/contact-us) to talk to us about building the human capacity around your AI-enabled workforce. ## Frequently asked questions **What is an AI workforce?** An AI workforce is a team of people and AI agents organised around the same work. Agents carry out defined, repeatable, multi-step tasks inside a business's systems, while people own judgement, relationships, consequential decisions, exceptions and accountability. Each agent has a human owner and a route for escalating what it cannot handle. **What is the agentic layer in a business?** The agentic layer is the set of AI agents working across a business's processes, together with what they need to run safely: access to data and systems, the rules and limits they work within, and the people who own, check and improve them. **How do you decide which work goes to an AI agent and which to a person?** Give an agent a workflow when the work is repetitive, rules-based, high-volume, system-driven and measurable. Keep judgement, relationships, decisions with financial, legal or customer consequences, and accountability with people. Where a workflow meets only some of those tests, use the agent to prepare the work and a person to finish it. **What roles are needed to manage AI agents?** Business analysts design the workflows; AI and machine learning engineers and data engineers build and connect the agents; data quality specialists and data analysts keep the data right; QA engineers test output; site reliability engineers and application support analysts run and monitor; customer operations staff handle exceptions; SOC analysts and security engineers watch agent access. **Will AI agents reduce headcount?** The evidence so far covers AI in general rather than agents specifically, and it shows little change for most organisations. In McKinsey's State of AI 2026, 14% of respondents from organisations using AI said AI had contributed to a fall in total headcount over the past year, and two-thirds reported little or no change. We advise planning for a change in the mix of work, with more people in exceptions, quality, data and improvement, and deciding in advance where freed capacity goes. **Can an offshore team support an AI agent workforce?** Yes, where the work around the agents is continuing and the team works inside your systems under your management. We build embedded teams for data quality, data engineering, testing, monitoring, exception handling and security. Plan data transfer safeguards before the team starts. Sources: [McKinsey, The State of AI in 2026: On the road to ROI (25 August 2026)](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai); [Gartner, Worldwide AI spending forecast (16 September 2026)](https://www.gartner.com/en/newsroom/press-releases/2026-09-16-gartner-forecasts-worldwide-ai-spending-to-grow-49-point-5-percent-in-2026); [Gartner, Agentic AI Fails Where Governance Stops (24 September 2026)](https://www.gartner.com/en/articles/agentic-ai-infrastructure-governance); [OECD, Empowering SMEs in the age of AI (April 2026)](https://www.oecd.org/content/dam/oecd/en/publications/reports/2026/04/empowering-smes-in-the-age-of-ai_7f58652c/bf5a9816-en.pdf); [Stanford HAI, AI Index Report 2026 (April 2026)](https://hai.stanford.edu/ai-index/2026-ai-index-report); [Wang et al., Agentic AI and Human-in-the-Loop Interventions: Field Experimental Evidence from Alibaba's Customer Service Operations (May 2026, working paper)](https://arxiv.org/abs/2605.14830); [Shao et al., Future of Work with AI Agents, Stanford (June 2025, preprint)](https://arxiv.org/abs/2506.06576); [Microsoft, 2026 Work Trend Index (5 May 2026)](https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization); [PwC, 2026 Global AI Jobs Barometer (15 June 2026)](https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-jobs-barometer.html); [Demirer, Musolff and Yang, Writing Code vs. Shipping Code, NBER Working Paper 35275 (May 2026, revised September 2026)](https://www.nber.org/papers/w35275); [ICO, A brief guide to international transfers (January 2026)](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/international-transfers/a-brief-guide-to-international-transfers/). Scroll [![footer-logo](https://blog.potentiam.co.uk/hubfs/footer-logo.svg)](https://blog.potentiam.co.uk/) We strategically build offshore teams that integrate seamlessly into your business. ##### Solutions - [Technology](https://www.potentiam.co.uk/solutions/technology) - [Finance & Accounting](https://www.potentiam.co.uk/solutions/finance-accounting) - [Marketing & Design](https://www.potentiam.co.uk/solutions/marketing-and-design) - [Data Analytics](https://www.potentiam.co.uk/solutions/data-analytics) - [Sales & Customer Success](https://www.potentiam.co.uk/solutions/sales-customer-service-solutions) - [Energy Consulting](https://www.potentiam.co.uk/solutions/energy-consulting) ##### Locations - [Cape Town](https://www.potentiam.co.uk/locations/south-africa) - [Johannesburg](https://www.potentiam.co.uk/locations/johannesburg) - [Bangalore](https://www.potentiam.co.uk/locations/india) - [Iași](https://www.potentiam.co.uk/locations/romania) - [São Paulo](https://www.potentiam.co.uk/locations/brazil) ##### Quick links - [About us](https://www.potentiam.co.uk/about-us) - [Careers](https://www.potentiam.co.uk/careers) - [Contact us](https://www.potentiam.co.uk/contact-us) © 2026 Potentiam. All rights reserved. [Privacy Policy](https://blog.potentiam.co.uk/privacy) ```json { "@context" : "https://schema.org", "@id" : "https://blog.potentiam.co.uk/how-to-build-an-ai-workforce#article", "@type" : "Article", "author" : { "@id" : "https://www.potentiam.co.uk/#organization" }, "dateModified" : "2026-11-09", "datePublished" : "2026-11-09", "description" : "How to build an AI workforce: which work AI agents should own, where people stay essential, and the roles, controls and team you need around the agents.", "headline" : "How to Build an AI Workforce of People and Agents", "image" : "https://5804539.fs1.hubspotusercontent-eu1.net/hubfs/5804539/blog/images/how-to-build-an-ai-workforce-hero-20260930-165319.webp", "keywords" : [ "AI workforce", "agentic workforce", "AI agents for business", "agentic layer", "human in the loop AI", "digital workforce" ], "mainEntityOfPage" : { "@id" : "https://blog.potentiam.co.uk/how-to-build-an-ai-workforce" }, "publisher" : { "@id" : "https://www.potentiam.co.uk/#organization" }, "wordCount" : 2951 } ``` ```json { "@context" : "https://schema.org", "@id" : "https://www.potentiam.co.uk/#organization", "@type" : "Organization", "address" : { "@type" : "PostalAddress", "addressCountry" : "GB" }, "description" : "Potentiam is a London-based strategic offshoring consultancy that helps mid-sized companies build high-performing, embedded offshore teams across five hubs in South Africa, Romania, India and Brazil: Cape Town, Johannesburg, Iași, Bengaluru and São Paulo.", "logo" : { "@type" : "ImageObject", "url" : "https://www.potentiam.co.uk/hubfs/logo.svg" }, "name" : "Potentiam Ltd", "sameAs" : [ "https://www.linkedin.com/company/potentiam-ltd" ], "url" : "https://www.potentiam.co.uk" } ``` ```json { "@context" : "https://schema.org", "@id" : "https://blog.potentiam.co.uk/how-to-build-an-ai-workforce#breadcrumb", "@type" : "BreadcrumbList", "itemListElement" : [ { "@type" : "ListItem", "item" : "https://www.potentiam.co.uk", "name" : "Home", "position" : 1 }, { "@type" : "ListItem", "item" : "https://blog.potentiam.co.uk", "name" : "Blog", "position" : 2 }, { "@type" : "ListItem", "name" : "How to Build an AI Workforce of People and Agents", "position" : 3 } ] } ``` ```json { "@context" : "https://schema.org", "@id" : "https://blog.potentiam.co.uk/how-to-build-an-ai-workforce#faq", "@type" : "FAQPage", "mainEntity" : [ { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "An AI workforce is a team of people and AI agents organised around the same work. Agents carry out defined, repeatable, multi-step tasks inside a business's systems, while people own judgement, relationships, consequential decisions, exceptions and accountability. Each agent has a human owner and a route for escalating what it cannot handle." }, "name" : "What is an AI workforce?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "The agentic layer is the set of AI agents working across a business's processes, together with what they need to run safely: access to data and systems, the rules and limits they work within, and the people who own, check and improve them." }, "name" : "What is the agentic layer in a business?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Give an agent a workflow when the work is repetitive, rules-based, high-volume, system-driven and measurable. Keep judgement, relationships, decisions with financial, legal or customer consequences, and accountability with people. Where a workflow meets only some of those tests, use the agent to prepare the work and a person to finish it." }, "name" : "How do you decide which work goes to an AI agent and which to a person?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Business analysts design the workflows; AI and machine learning engineers and data engineers build and connect the agents; data quality specialists and data analysts keep the data right; QA engineers test output; site reliability engineers and application support analysts run and monitor; customer operations staff handle exceptions; SOC analysts and security engineers watch agent access." }, "name" : "What roles are needed to manage AI agents?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "The evidence so far covers AI in general rather than agents specifically, and it shows little change for most organisations. In McKinsey's State of AI 2026, 14% of respondents from organisations using AI said AI had contributed to a fall in total headcount over the past year, and two-thirds reported little or no change. We advise planning for a change in the mix of work, with more people in exceptions, quality, data and improvement, and deciding in advance where freed capacity goes." }, "name" : "Will AI agents reduce headcount?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Yes, where the work around the agents is continuing and the team works inside your systems under your management. We build embedded teams for data quality, data engineering, testing, monitoring, exception handling and security. Plan data transfer safeguards before the team starts." }, "name" : "Can an offshore team support an AI agent workforce?" } ] } ```