U.S. SMID-cap and global application software Report Interpretation

Overview

The report examines whether AI-driven productivity will shrink white-collar employment and undermine software priced by user seat. Bernstein concludes that firm-level evidence currently points in the opposite direction: companies adopting AI most intensively are hiring faster, suggesting that productivity gains are often being reinvested in growth rather than captured mainly through headcount reductions and wider margins.

Core views

Bernstein starts with a central concern for application software: AI agents and copilots are increasingly performing work formerly associated with developers, junior employees and support functions. Coding copilots are helping teams ship software faster, while desktop and customer-service agents automate reporting and self-service tasks. The report cites ServiceNow Agent Assist adoption at roughly one-quarter of its customer base and Sierra's nearly $16 billion private-market valuation as illustrations of real-world agent adoption. If these tools reduce employee numbers, vendors that charge per user could face pressure even as they experiment with alternatives such as token consumption, infrastructure-based meters and return-on-investment pricing. The report nevertheless argues that enterprise buyers retain rational reasons to prefer seats. Headcount is a knowable measure of organizational capacity, making seat-based costs easier to forecast and budget than potentially volatile consumption. Application software also creates value largely by improving employee productivity, so scaling price with employees retains a value-based logic. Bernstein therefore expects seats to remain an important component of software sales, even when vendors combine them with alternative meters and even though some customers or workloads will not suit seat pricing. The empirical centerpiece is Ramp research using customers' actual finance and accounting spend to identify AI usage. Adoption begins with the first month of the earliest three-month period in which a company spends at least $100 with AI vendors every month; intensity is measured as AI spending per baseline employee over the first three post-adoption months. Heavy adopters broadly use coding agents, API-token-based internal agents and AI-native software, while light adopters mainly deploy chat tools for search, synthesis and creative work. The sample ends in February 2026 and therefore does not include the additional technology-hiring momentum Bernstein says appeared by mid-year. On this measure, U.S. companies applying AI deeply hired more aggressively than like-for-like low and non-adopters. The high-adoption employment increase was statistically significant, reaching 99% confidence for some results. Ramp largely compared the same peer-company sets before and after adoption rather than contrasting high-growth technology firms with slower industries. Heavy adopters grew more than 10% faster than the control and lower-adoption groups on average. Companies observed 24 months after initial adoption were more than 50% larger than peers with lower or no adoption, although fewer early adopters are available at that horizon, widening the error bars and making the midpoint less precise. The robust conclusion is that high adopters grew more, not that the exact 24-month magnitude is certain. The job composition also pushes back against a simple displacement narrative. Engineers and sales were strong, as Bernstein expected for growth-oriented front-office functions, but administrative and customer-service employment also increased. Hiring was strong among workers with bachelor's degrees and especially at entry level, contrary to the idea that AI primarily eliminates junior roles. PhD employment was more likely to be flat at leading adopters, while low adopters showed greater demand for advanced-degree and scientist roles; one possible explanation offered is that high adopters already employed more PhDs before adoption. Most functions grew more slowly than total employment, implying genuine productivity gains, but the overall company-growth effect still generated positive labor requirements. Finance, scientists and operations were the three functions with much weaker or no statistically significant positive growth, and human resources was not separately reported. Adoption itself is already substantial in Ramp's sample. More than 40% of companies were at least light adopters by January 2026, well above the high-teens rate in the contemporaneous U.S. Census survey. Bernstein says the gap could reflect different definitions, respondent awareness or Ramp's tilt toward technology-forward customers. Adoption was highest in the NAICS Information category at approximately 54%, followed by Finance and Insurance at approximately 44% and Professional, Scientific and Technical Services at 36%; most other industries were at 11% to 22%. Separately, U.S. software-development job postings on Indeed had begun recovering from their early-2025 trough, which Bernstein presents as corroborating evidence beyond Ramp's customer base. Bernstein explains the employment result through corporate incentives. It argues that companies often receive more valuation benefit from sustained growth acceleration than from a one-time margin step-up. Higher productivity can therefore be used to retain or expand a more productive workforce, create products, increase selling capacity and accelerate revenue rather than simply remove employees. Around two-thirds of developers work in industries the report considers more likely to reinvest in product-led growth, while only approximately 25% work in industries more likely to shrink teams. The likely growth-oriented group extends beyond technology and internet companies to banks and selected automotive, aerospace and defense manufacturers, as well as individual technology-intensive leaders in other industries. Historical evidence supports this mechanism. Despite major productivity gains between 1989 and 2019, like-for-like S&P 500 EBIT margins expanded by only about 1.5%, or roughly 150 basis points, over 30 years. Bernstein argues that sustained excess margins are difficult to preserve because competitive forces tend to erode them, while revenue growth at stable margins adds incremental profit and earnings per share without the same diminishing room for further margin expansion. During the mature PC era of the 1990s, sharply higher white-collar productivity coincided with strong U.S. GDP growth and increased professional and business-services employment rather than broad white-collar contraction; the internet era subsequently reinforced productivity in data and information industries. GenAI may similarly ease constraints on identifying, building and selling new offerings, making organic growth easier to pursue. The conclusion is conditional rather than universal. Dominant or mature businesses with limited growth opportunities may prefer margins, and companies lacking the products, skills, resources or sales capacity to turn AI into growth may allow savings to reach the bottom line or invest outside software. Accordingly, seat-based products aimed at finance, scientific or operations functions may receive less employment support than products serving faster-growing roles. The evidence is also too early to establish how jobs and demand ultimately evolve. Even with these caveats, Bernstein views productivity-induced hiring as a meaningful counterweight to fears that AI will make seat-based software licensing structurally obsolete. The 31 August 2026 ticker table leaves models and recommendations unchanged. Ratings are Outperform for Atlassian, GitLab, Okta, ServiceNow, Microsoft, SAP, Adobe and Workday; Market-Perform for Zoom and HubSpot; and Underperform for Salesforce. Current prices and targets are respectively USD 194.17 and USD 309.00 for TEAM, USD 46.54 and USD 60.00 for GTLB, USD 173.04 and USD 143.00 for OKTA, USD 147.99 and USD 248.00 for NOW, USD 96.68 and USD 108.00 for ZM, USD 507.29 and USD 660.00 for MSFT, USD 257.54 and USD 195.00 for CRM, USD 220.89 and USD 319.00 for SAP, USD 292.79 and USD 379.00 for ADBE, USD 197.45 and USD 238.00 for WDAY, and USD 261.36 and USD 220.00 for HUBS. The table reports relative-performance figures of (9.8)%, (22.1)%, 67.6%, (38.3)%, (0.2)%, (18.9)%, (18.5)%, (37.8)%, (36.9)%, (33.4)% and (64.9)% in the same order; these are reported historical relative-performance figures rather than implied target-price upside.

Analysis framework

Bernstein first frames the risk that AI automation poses to employee-based software pricing, then tests that concern using Ramp's firm-level spending and workforce study. It compares employment before and after adoption within like-for-like company sets, examines differences by job function and education level, checks adoption rates by industry, and uses Indeed employment data as outside confirmation. The report then connects the findings to historical PC- and internet-era productivity cycles and to the corporate choice between reinvesting efficiency in growth or taking it as margin.

Methodology notes

• Other

  Pre- and post-adoption comparison within like-for-like company sets

  Ramp compares similar companies, largely against their own employment paths before and after intensive AI adoption, to reduce the bias that would arise from simply comparing fast-growing technology companies with slower industries.

• Industry AnalysisVolume-price decomposition

  Employment growth by function, education and adoption intensity

  The report separates aggregate headcount growth into roles and worker categories to determine whether overall hiring masks losses in administrative, customer-service, junior or other groups.

• Cycle and Business ConditionsBusiness-Cycle Inflection Analysis

  Hiring-trend confirmation

  Bernstein compares the Ramp results with the recovery in U.S. software-development job postings from the early-2025 trough to assess whether the employment signal extends beyond Ramp's sample.

• Other

  Historical PC- and internet-era productivity analogy

  The report uses earlier technology-driven productivity booms to argue that efficiency can be reinvested in faster growth and additional knowledge-worker employment rather than primarily in sustained margin expansion.

• Valuation methodsP/E and PEG Valuation

  Adjusted earnings multiples

  The ticker table compares covered companies using adjusted earnings per share and adjusted P/E estimates for 2026A, 2027E and 2028E.

• Valuation methods

  EV/Sales valuation for HubSpot

  HubSpot is shown using EV/Sales rather than adjusted P/E, reflecting the report's stated company-specific valuation convention.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Atlassian (TEAM)

  Covered application-software company rated Outperform; the report leaves its model, USD 309.00 target and recommendation unchanged.

  Strengths

  Potential participation in resilient seat-based demand if AI-driven productivity supports product-team hiring.

  Comparison

  Included in Bernstein's multi-company software ticker table.

  Risks

  The report's sector thesis remains dependent on early employment data and may vary by customer function.

• GitLab (GTLB)

  Covered software company rated Outperform; the report leaves its model, USD 60.00 target and recommendation unchanged.

  Strengths

  Potential exposure to growth in AI-enabled software-development teams.

  Comparison

  Included in Bernstein's multi-company software ticker table.

  Risks

  The report does not establish that observed aggregate hiring trends will translate uniformly into company-level seat growth.

• Okta (OKTA)

  Covered software company rated Outperform; the report leaves its model, USD 143.00 target and recommendation unchanged.

  Strengths

  Potential support from broader organizational hiring under Bernstein's sector thesis.

  Comparison

  Included in Bernstein's multi-company software ticker table.

  Risks

  The employment evidence is early and uneven across roles.

• ServiceNow (NOW)

  Covered application-software company rated Outperform; the report leaves its model, USD 248.00 target and recommendation unchanged.

  Strengths

  Agent Assist adoption at roughly one-quarter of the customer base illustrates practical AI-agent demand.

  Weaknesses

  Support-oriented automation can also raise concern about displacement of customer-service seats.

  Comparison

  Used as a report example of enterprise AI-agent adoption.

  Risks

  Automation benefits may not result in additional seats in every function.

• Zoom Video Communications (ZM)

  Covered software company rated Market-Perform; the report leaves its model, USD 108.00 target and recommendation unchanged.

  Strengths

  Potential participation in continued employee-based enterprise software demand.

  Comparison

  One of two Market-Perform names in the ticker table.

  Risks

  The sector-wide hiring thesis does not establish company-specific operating benefits.

• Microsoft (MSFT)

  Covered software company rated Outperform; the report leaves its model, USD 660.00 target and recommendation unchanged.

  Strengths

  The report references Microsoft results and desktop copilots in discussing adoption and productivity.

  Comparison

  Cited alongside HubSpot and Salesforce in the report's discussion of front-office adoption.

  Risks

  The report notes that current company commentary does not yet clearly demonstrate improved front-office adoption.

• Salesforce (CRM)

  Covered application-software company rated Underperform; the report leaves its model, USD 195.00 target and recommendation unchanged.

  Strengths

  Could benefit if AI productivity is reinvested in sales and other growth functions.

  Weaknesses

  The company-specific recommendation remains Underperform despite the more constructive sector-level seat thesis.

  Comparison

  The only Underperform name in the ticker table.

  Risks

  The report notes that current results and commentary do not clearly point to improved front-office adoption.

• SAP (SAP)

  Covered global software company rated Outperform; the report leaves its model, USD 319.00 target and recommendation unchanged.

  Strengths

  Potential support from enterprises retaining predictable employee-linked pricing.

  Comparison

  Included in Bernstein's global software coverage table.

  Risks

  Customer industries with limited growth opportunities may direct AI savings to margins rather than hiring.

• Adobe (ADBE)

  Covered software company rated Outperform; the report leaves its model, USD 379.00 target and recommendation unchanged.

  Strengths

  Potential exposure to growth-oriented reinvestment of AI productivity.

  Comparison

  Included in Bernstein's multi-company software ticker table.

  Risks

  The report does not quantify a company-specific seat benefit.

• Workday (WDAY)

  Covered application-software company rated Outperform; the report leaves its model, USD 238.00 target and recommendation unchanged.

  Strengths

  Potential support if overall enterprise employment continues to expand.

  Weaknesses

  Human-resources functions were not separately identified in the underlying study.

  Comparison

  Included in Bernstein's multi-company software ticker table.

  Risks

  Limited visibility into HR hiring prevents a direct read-through from the study.

• HubSpot (HUBS)

  Covered application-software company rated Market-Perform; the report leaves its model, USD 220.00 target and recommendation unchanged.

  Strengths

  Potential exposure to hiring in sales and other growth-oriented functions.

  Comparison

  Valued using EV/Sales rather than the adjusted P/E convention used for most table entries.

  Risks

  The report says current company results do not clearly point to improved front-office adoption.

Key data

• Heavy-adopter employment advantage>10%Average growth above the control non-adopters and lower-adoption peers.
• Twenty-four-month company-size difference>50%High adopters were more than 50% larger than lower or non-adopting peers, but the error bar widened at this horizon.
• Statistical confidenceUp to 99%Confidence that high adopters experienced greater growth, depending on the measured data point.
• AI adoption thresholdAt least $100 per month for three consecutive monthsThe first month of the earliest qualifying period defines adoption.
• Ramp sample end dateFebruary 2026The sample does not include the further technology-hiring momentum Bernstein observed into mid-year.
• Companies with any AI adoption>40% by January 2026Ramp's rate was well above the high-teens result from the contemporaneous U.S. Census survey.
• AI adoption by leading industriesInformation ~54%; Finance/Insurance ~44%; Professional/Scientific/Technical Services 36%Most other industries recorded adoption rates of 11% to 22%.
• Developer employment mix~2/3 growth-oriented; ~25% more likely to shrink teamsBernstein's industry classification of where developers are employed.
• Long-run like-for-like EBIT margin expansion1.5% (~150 bps)S&P 500 comparison for 1989 to 2019 despite substantial labor-productivity growth.
• TEAM — AtlassianOutperform; current USD 194.17; target USD 309.00; relative performance (9.8)%Adjusted EPS 2026A/2027E/2028E: USD 5.85/7.61/9.35; adjusted P/E: 33.2x/25.5x/20.8x.
• GTLB — GitLabOutperform; current USD 46.54; target USD 60.00; relative performance (22.1)%Adjusted EPS 2026A/2027E/2028E: USD 0.96/1.00/1.34; adjusted P/E: 48.2x/46.6x/34.7x.
• OKTA — OktaOutperform; current USD 173.04; target USD 143.00; relative performance 67.6%Adjusted EPS 2026A/2027E/2028E: USD 3.50/4.05/4.45; adjusted P/E: 49.4x/42.7x/38.9x.
• NOW — ServiceNowOutperform; current USD 147.99; target USD 248.00; relative performance (38.3)%Adjusted EPS 2026A/2027E/2028E: USD 3.50/3.92/4.69; adjusted P/E: 42.2x/37.8x/31.5x; stated base year is 2025.
• ZM — ZoomMarket-Perform; current USD 96.68; target USD 108.00; relative performance (0.2)%Adjusted EPS 2026A/2027E/2028E: USD 5.92/6.41/6.94; adjusted P/E: 16.3x/15.1x/13.9x.
• MSFT — MicrosoftOutperform; current USD 507.29; target USD 660.00; relative performance (18.9)%Adjusted EPS 2026A/2027E/2028E: USD 17.28/19.96/24.01; adjusted P/E: 29.4x/25.4x/21.1x.
• CRM — SalesforceUnderperform; current USD 257.54; target USD 195.00; relative performance (18.5)%Adjusted EPS 2026A/2027E/2028E: USD 12.52/17.13/15.49; adjusted P/E: 20.6x/15.0x/16.6x.
• SAP — SAPOutperform; current USD 220.89; target USD 319.00; relative performance (37.8)%Adjusted EPS 2026A/2027E/2028E: EUR 6.10/7.71/8.72; adjusted P/E: 31.2x/24.7x/21.8x; stated base year is 2025.
• ADBE — AdobeOutperform; current USD 292.79; target USD 379.00; relative performance (36.9)%Adjusted EPS 2026A/2027E/2028E: USD 20.95/24.99/29.15; adjusted P/E: 14.0x/11.7x/10.0x; stated base year is 2025.
• WDAY — WorkdayOutperform; current USD 197.45; target USD 238.00; relative performance (33.4)%Adjusted EPS 2026A/2027E/2028E: USD 9.23/11.58/15.30; adjusted P/E: 21.4x/17.1x/12.9x.
• HUBS — HubSpotMarket-Perform; current USD 261.36; target USD 220.00; relative performance (64.9)%2026A/2027E/2028E figures are USD 9.70/13.20/15.91; EV/Sales multiples are 3.8x/3.2x/2.8x; stated base year is 2025.

Impact & implications

The report says the evidence reduces, but does not eliminate, the structural concern that AI will erode application-software revenue tied to employee seats. If intensive adoption raises productivity and companies reinvest that capacity in growth and hiring, seat counts can continue expanding while hybrid seat, consumption and outcome-based pricing evolves. The benefit is likely to vary by customer industry and employee function, with weaker support for products aimed mainly at finance, scientific or operations roles.

Risks

• The Ramp evidence may still be too early to show how long-term labor demand and job design will evolve.
• Error bars widen at longer horizons because relatively few companies adopted AI 24 months earlier.
• Ramp's customer base may skew toward technology-forward buyers, limiting the representativeness of its adoption rates.
• Differences between Ramp's adoption rate and the U.S. Census survey may reflect inconsistent definitions or respondent awareness.
• Finance, scientist and operations roles showed much weaker or no statistically significant positive growth, reducing potential support for software aimed at those functions.
• The study did not separately report functions such as human resources, leaving important areas of application-software demand uncertain.
• Mature companies, businesses with limited organic opportunities and industries where technology is mainly operational may take AI savings as margin or invest them outside software rather than expand hiring.

What to watch

• Whether heavy AI adopters continue to outpace low and non-adopters in employment as the sample extends beyond February 2026.
• Whether the recovery in U.S. software-development job postings continues after the early-2025 trough.
• Hiring trends by function, particularly finance, scientists, operations and the unreported human-resources category.
• Whether AI adoption broadens beyond Information, Finance and Insurance, and Professional, Scientific and Technical Services.
• How application-software vendors balance seats with consumption, infrastructure-based and outcome-based pricing.