Redesign the Work, Don't Bolt AI onto Old Jobs

Staff at companies that redesign work report more from AI than staff at companies that only add tools, and that makes L&D's job bigger than tool training. BCG's June 2026 AI at Work report says "most companies still treat AI as a tool for individual productivity" and urges redesigning a few core processes end to end. 3

Among 10,990 respondents who placed their own employer in BCG's tiers, 67% at companies that redesign workflows or build new business models report measurable business improvement, against 43% at companies that only deploy tools. Across all 11,749 respondents, 88% believe they need major upskilling in the next five years, and 36% feel properly trained. 3

BCG's interviews at more than 50 AI front-runner firms find almost all the large ones building new AI-era roles, and two-thirds of the firms describing training so more people can direct AI, not just use it. 1

None of this proves redesign causes the payoff; redesigners may already be ahead. It does put L&D's decision in view: who learns to take a workflow apart, and who teaches them.

1. 1. Reported value sits with the redesign more than the tools. Staff who place their employer among the redesigners report measurable business improvement more often than staff at deploy-only employers, 67% against 43%. 3
2. 2. Front-runners are creating roles whose job is to run AI work. BCG names five: citizen builders, workflow designers, AI guardians, domain anchors, and agent shepherds. 1
3. 3. Front-runners say they train people to direct AI; only about a third of surveyed workers feel properly trained. Two-thirds of the front-runner firms BCG interviewed describe that training; in BCG's worker survey, 36% feel properly trained. 13
4. 4. Most of the AI impact users report goes with organizational factors, and few say reinvention is rewarded. In Microsoft's survey of AI users, organizational factors account for 67% of reported AI impact, and 13% say they're rewarded for reinvention of work with AI even if results aren't met. 4
5. 5. The return premium belongs to AI-adoption leaders and is smaller than BCG's September headline. The 6% of 600-plus large US public companies rated AI leaders ran 9.3 points above the sample median in industry-adjusted return over three years; BCG's "more than 11 percentage points" a year is the leader-laggard spread. 12

Brief 018 asked what an existing role is for once AI absorbs its routine work. Brief 025 asked how new hires learn judgment when the starter tasks go, and Brief 004 asked whether reskilling can keep pace with deployment. This Brief moves from the role to the workflow: who takes a process apart, decides what agents do, and runs the result, and what L&D must build for that.

The causal case is open. Reported gains sit where respondents say work was redesigned, the front-runner interviews show the roles and training built around that work, and the return premium sits with a small group of AI-adoption leaders. In the strongest counter-evidence, 9 in 10 executives in an NBER survey report no AI impact on employment or productivity over three years. 1235

Reported value sits with redesign, and returns with AI leaders

BCG's AI at Work survey, published 3 June 2026, has respondents place their employer in Deploy (adopting generative AI tools), Reshape (redesigning end-to-end workflows), or Invent (building new business models and products). 3

Against Deploy-only companies, respondents at Reshape or Invent companies are more likely to see measurable business improvement (67% against 43%), take part in process redesign (43% against 12%), and go through a major reskilling initiative (65% against 47%). 3

Among 9,923 regular AI users, the share reporting measurable impact is 55% with limited strategic clarity and limited tool access, 60% with strong tool access but limited clarity, 80% with strong strategic clarity but limited tool access, and 83% with both. Strong clarity means AI is a priority with a clear strategy and guidelines for time saved. BCG's bar for strong tool access is low: access to tools or to testing. 3

The BCG Institute sorted more than 600 large US public companies into four AI-adoption tiers; 6% qualify as leaders. Over three years, leaders ran 9.3 points above the sample median in industry-adjusted total shareholder return, the active tier ran 0.6 points above it, and laggards ran 1.7 points below it. The study describes the change as "rethinking workflows end-to-end rather than bolting AI onto existing processes," but its tiers rank AI adoption, it does not say whether the figures are annualized, and the result is a correlation. 2

The new roles design, govern, and supervise the work

BCG's 24 September article, drawing on a qualitative study of more than 50 front-runner firms, describes five new roles. Citizen builders build their own apps and agents inside the business. Workflow designers rethink multistep processes end to end, deciding what AI absorbs and what stays human. AI guardians govern risk and compliance. Domain anchors encode expert knowledge in reusable form. Agent shepherds oversee a fleet of agents: how they run, what they can access, and when they are shut down. 1

Read together, the five cover the life of a redesigned workflow. They are new roles, not what remains of an old job, which was Brief 018's question. 1

In BCG's June survey, 47% of respondents agree their roles have shifted toward managing and directing AI (30% disagree), and half of the regular AI users who know about agents say their company has no clear guidance on managing teams of people and AI. 3

Teaching the tool is the smaller job

BCG tells traditional companies to rewire talent and performance processes, judging people by the impact they create and the judgment behind it. 1

BCG's training numbers have not moved since 2025, and 66% of frontline regular AI users get limited or no guidance on what to do with the time AI saves. This Brief reads that as a design problem more than a supply problem: people are trained on tools for work nobody rebuilt. 3

In Microsoft's 2026 Work Trend Index, a vendor survey of 20,000 AI-using knowledge workers, organizational factors such as culture, manager support, and talent practices account for 67% of the AI impact workers report, against 32% for individual mindset and behavior. Yet 45% say it feels safer to focus on current goals than to redesign work with AI. 4

Autonoma forecast: Over the next 12 months, companies that keep adding AI tools to unchanged processes will keep trailing redesigners on reported business improvement, while front-runners keep adding roles that design, govern, and supervise AI work. L&D programs built only on tool fluency will fall behind that demand. 13

The first five track the sources. The last two are local checks.

• BCG's next AI at Work edition shows the Reshape or Invent advantage holding, or closing. 3
• The 36% who feel properly trained moves, or stays flat a third year. 3
• Workflow designer, AI guardian, and agent shepherd duties appear in incumbents' job architectures. 1
• The 13% who say reinvention is rewarded even if results aren't met starts to rise. 4
• Research tests whether redesign, not prior performance, explains the leaders' return gap. 25
• Your L&D catalog teaches at least one team to redesign a workflow, not only to use a tool.
• Each workflow on your AI plan has a named owner for its redesign, guardrails, and agents.

For CLOs and L&D: The process owner owns the redesign; L&D owns the capacity to do it. Fund that capacity beside the tool courses: teach teams to map a workflow, decide what AI absorbs and what stays human, and move the handoffs. Count who took part in a redesign. 13

For CHROs: Test BCG's five roles against your own workflows, then put the ones that fit into the job architecture with owners, career paths, and assessment. 1

For business and function leaders: With 45% of AI users finding it safer to stick with current goals than to redesign work, the incentive problem is yours before it is L&D's. 4

For CFOs and boards: Ask for redesign measures beside adoption counts, and quote the premium as 9.3 points over the median across three years for the 6% ranked AI leaders, a correlation. 2

For CIOs: Every tool rollout should name the workflow it changes and who redesigns it. 3

Weight: Strong. The strongest objection is selection: the firms that redesign may have been ahead before they did. In NBER Working Paper 34836, a survey of nearly 6,000 executives in four countries, 69% of firms actively use AI, with higher use at younger and more productive firms, and 9 in 10 executives report no impact of AI on employment or productivity over the past three years. If stronger firms use AI more, they may also be the ones that redesign. 5

BCG's labels look generous too: 8,071 of its 10,990 respondents put their employer in Reshape or Invent, though BCG says most companies still treat AI as an individual tool. The tiers may track how much AI a company uses as much as whether it redesigned. Microsoft surveyed only AI users, the front-runners were picked because they lead, and three of the five sources are BCG's. 134

Most of that is right; this Brief does not rest on cause. The case for acting is that front-runners are building these roles now and that half of regular AI users who know about agents lack clear guidance on managing teams of people and AI. A study showing that tools on unchanged processes deliver the same gains, or that the gap vanishes once prior performance is controlled, would cut this Brief back to the roles alone; none of these sources runs that test. 3

Many AI training plans are a license, a course on prompts, and a help channel. They teach people to use a tool inside a process nobody changed. BCG's front-runners take the workflow apart instead, decide what the agents do and where a person steps in, and name someone to watch the agents.

Ask four questions of each workflow on the AI plan. Who redesigns it? Who keeps it in bounds? Who supervises the agents in it? Who turns the expert's knowledge into something the next person can reuse? Each blank answer is a role to build and a curriculum to write.

If every answer points back to the vendor's onboarding course, the company is bolting AI onto old jobs.