# salesforce-tallapragada-how-engineering-became-agentic-2026-05-27

## Veille

Official **Salesforce News** blog post (*Agentic Enterprise* section, *"Pioneering the Agentic Shift Within Salesforce Engineering"* series), published on **May 27, 2026** (6-minute read) by **Srinivas "Srini" Tallapragada**, *President and Chief Engineering and Customer Success Officer* at Salesforce. Direct follow-up to an earlier post (*"How we got our engineers to use AI — without breaking everything"*) which recounted crossing **>90% adoption**. **Pivot thesis**: Salesforce Engineering moved from a world where AI was a useful *copilot* to one where **agentic tools drive the software development lifecycle (SDLC) itself** — writing code, reviewing PRs, generating tests, updating documentation, managing deployments, coordinating work once handled through human handoffs. **Canonical signal decision**: org-wide standardization on **Claude Code** + ***"we removed all token limits"*** — *"remove every last piece of friction between our engineers and the tools that make them faster and more effective"*. **Major empirical result** (April 2026 vs April 2025): work items completed per developer **+50.8%**, PRs merged per developer **+79%**, and above all **Effective Output score** (an ML measure of the **real value of delivered code**, not volume) **+151.3% year over year**. **Flagship use case**: migration of **33 API endpoints** to a cloud-native architecture, estimated at **~231 person-days** (7 per API) the traditional way, completed in **13 days — 18× faster** — via a **rule-based framework built in Claude** (markdown files + reference implementations), with PR feedback continuously fed back into the rule set, **autonomous LLM loops (build, fix, validate)** with no manual intervention, parallelized across isolated environments → **5 PRs**, the largest delivering **21 endpoints with 100% test coverage**. **No speed↔quality tradeoff**: through the **Engineering 360** platform (centralizing engineering data from hundreds of systems), **total incidents drop by 5%** despite the rise in PRs (*"quality doesn't suffer from speed. It benefits from it"*), thanks to **security guardrails and quality standards structurally embedded** in the agentic workflow (Trust as the #1 value). **SDLC overhaul**: once AI is adopted, engineers **tear down and rebuild** workflows (which processes to eliminate? which handoffs are now unnecessary? where does a human still do work an agent could own?). **New engineering craft**: **Claude Code skills** (packaged, reusable capabilities encoding team context, naming conventions, patterns) become a shared, composable **engineering artifact**; **AI Expert Suite** + **Salesforce Foundation Plugins** = an institutionalized, curated skills library (internal benchmark: **higher accuracy and reliability, reduced unnecessary cost**); **subagents & agent teams** parallelize workstreams (*"They describe the outcome, and a set of coordinated agents figures out the steps"*). **What remains hard**: (1) **context management** in long sessions — **CLAUDE.md file quality** varies widely and weighs heavily on output quality; (2) **agentic security** = a fundamentally different model (agents that *act*, not just *suggest* → increased blast radius); (3) **evolving roles** (how do juniors become seniors if AI absorbs entry-level work? role of the designer/PM? the execution unit = scrum team → experiments with 1- or 3-person units). Conclusion: *"It changed what was economically possible"*; the stated ambition is **"the most automated, agentic SDLC in the industry"**. Directly intersects with Gupta (*cost of a completed outcome*, marginal token utility), Greenwald/Sierra (outcome-based pricing), DORA (ROI / cost per feature) and the BFM/Girard debate (token as a value fuel, not a cost to cut).

## Titre Article

How Salesforce Engineering Became Truly Agentic

## Date

2026-05-27

## URL

https://www.salesforce.com/news/stories/how-engineering-became-agentic/

## Keywords

Agentic SDLC, agentic SDLC, Claude Code, removal of token limits, removed all token limits, remove friction, Effective Output score, real value of code, work items per developer +50, 8%, PRs merged per developer +79%, Effective Output +151, 3% YoY, migration 33 endpoints, 231 person-days, 18× faster, 13 days, rule-based framework Claude, reference implementations, autonomous LLM loops build fix validate, parallelization across isolated environments, 5 PRs, 21 endpoints 100% test coverage, Engineering 360, incidents -5%, no speed/quality tradeoff, quality benefits from speed, Trust as top value, embedded security guardrails, SDLC overhaul, tear down rebuild workflows, Claude Code skills, engineering artifact, AI Expert Suite, Salesforce Foundation Plugins, curated skills library, reducing unnecessary cost, subagents, agent teams, parallel workstreams, describe the outcome, context management, CLAUDE.md quality, agentic security, blast radius, evolving roles, junior senior, scrum team, 1- or 3-person units, economically possible, most automated agentic SDLC, 90% adoption, agentic FinOps, cost per outcome, Srinivas Tallapragada, Salesforce

## Authors

**Srinivas « Srini » Tallapragada** — *President and Chief Engineering and Customer Success Officer* de **Salesforce**. Plus d'une décennie chez Salesforce, dirige l'ingénierie mondiale de la plateforme unifiée. Auteur de la série *Agentic Enterprise* sur le blog Salesforce News ; ce billet (27 mai 2026) est la **suite** d'un premier opus consacré à l'adoption de l'IA par les milliers d'ingénieurs Salesforce (*« How we got our engineers to use AI — without breaking everything »*). Position d'autorité = **dirigeant exécutif** parlant en son nom et au nom d'une organisation d'ingénierie à grande échelle (donnée terrain à l'échelle d'un hyperscaler SaaS), avec accès aux métriques internes (Engineering 360, Effective Output).

## Ton

**Profile**: Executive leadership post (executive blog / *progress report*), first-person plural (*"we"*), aimed at engineering operators and leaders (CTOs, VPs of Engineering, EMs), practitioners, and, implicitly, recruitment (*"if you want to work on an AI-native engineering team"*). Register: **corporate-confident but data-backed**, **medium-high** technical level (assumes familiarity with PRs, SDLC, subagents, CLAUDE.md, reference implementations, test coverage) while remaining readable for a non-engineering decision-maker.

**Style**: Executive prose — measured claims, structured by action subheadings (*Ramping with Claude Code*, *What Agentic Transformation actually looks like*, *More output, better quality — at the same time*, *Rethinking the SDLC*, *Skills, subagents, and the new engineering craft*, *What we're still figuring out*, *The direction is clear*). Alternation of **narrative + figure**: a direction is announced (*"we removed all token limits"*), proven with data (+50.8%, +79%, +151.3%), illustrated with a single case (the 18× migration). **Measured honesty**: a whole section (*What we're still figuring out*) admits what remains hard (context, security, roles) — a register of transparency that reinforces credibility without undermining the message. No AGI overselling; the talk is of outcomes, quality, economics.

**Key aphorisms**:
- ***"We removed all token limits."*** (signal decision).
- ***"Remove every last piece of friction between our engineers and the tools that make them faster and more effective."***
- ***"When agentic tools get applied properly, quality doesn't suffer from speed. It benefits from it."***
- ***"They describe the outcome, and a set of coordinated agents figures out the steps."***
- ***"It changed what was economically possible."***
- ***"The engineering organization of the future doesn't look like the organization of today with AI bolted on. It looks fundamentally different."***

**Metaphors / frames at work**:
- ***Copilot → driver***: the shift from AI as assistant to AI that *drives* the SDLC.
- ***AI bolted on*** vs a **natively rethought organization** — the post's structuring antithesis.
- ***Removing friction*** as an investment philosophy: the token limit is not a cost safeguard but **friction** to be eliminated.
- ***Effective Output***: shifting the measure from *volume* (lines, PRs) to the **real value** of code — a direct echo of the *cost of a completed outcome*.

**Epistemic position**: an operator's account of experience at scale, backed by internal instrumentation (Engineering 360, ML score). To be read with the **usual caveat**: official communication from a vendor (Salesforce) about its own transformation, a partner of the tool being praised (Claude Code) — figures not third-party audited, cases hand-picked. But the **internal consistency** and the admission of difficulties make it a solid field source.

**Authority**: built from (a) **position** (President/Chief Engineering Officer of a SaaS hyperscaler), (b) **scale** (thousands of engineers, >90% adoption), (c) **proprietary data** (Effective Output, Engineering 360), (d) the **series** (an avowed progress report, continuity with the previous post), (e) the **concrete, quantified case** (the 18× migration).

## Pense-betes

- **Date / source**: **May 27, 2026**, official **Salesforce News** blog (*Agentic Enterprise* section), 6 min. Author: **Srini Tallapragada** (President & Chief Engineering and Customer Success Officer).
- **Follow-up to**: *"How we got our engineers to use AI — without breaking everything"* (>90% adoption crossed). This post = the **next step**: no longer adopting, but **rebuilding the SDLC**. ### The signal decision (core for the agentic FinOps slot)
- **Org-wide standardization on Claude Code** + ***"we removed all token limits"***.
- Stated logic: the token limit is **friction**, not a cost safeguard. *"Remove every last piece of friction."*
- ⚠️ **Direct counterpoint** to the "cut the token budget" reflex → intersects with Willenbrock (*"those cutting token budgets never got past the pilot stage… cost center instead of a capability"*) and Mollick. ### The numbers (April 2026 vs April 2025) | Metric | YoY change | |----------|---------------| | Work items completed / developer | **+50.8%** | | PRs merged / developer | **+79%** | | **Effective Output score** (real value, ML, not volume) | **+151.3%** | | Total incidents (despite ↑ PRs) | **−5%** |
- **Effective Output** = the real find: measuring the **value** of delivered code, not the volume → a cousin of the *cost of a completed outcome* (Gupta) and outcome-based pricing (Greenwald). ### The migration case (proof by example)
- **33 API endpoints** → cloud-native architecture. Traditional: **~231 person-days** (7/API). Completed in **13 days = 18×**.
- Recipe: **rule-based framework built in Claude** (markdown + reference implementations) → PR feedback **continuously fed back** into the rule set → **autonomous LLM loops (build, fix, validate)** with no intervention → **parallelization** across isolated environments.
- Output: **5 PRs**, the largest = **21 endpoints, 100% test coverage**. *"It changed what was economically possible."* ### The new craft
- **Claude Code skills** = an engineering artifact (team context, conventions, patterns) — **shared, composable**.
- **AI Expert Suite** + **Salesforce Foundation Plugins** = a curated library → internal benchmark: **+accuracy, +reliability, −unnecessary cost**.
- **Subagents / agent teams** → the engineer **describes the outcome**, coordinated agents find the steps (an end to context-switching across 5 systems).
- Top skill of 2026: **structuring a problem for an agentic system**, knowing **when to delegate vs stay in the loop**, **building reusable patterns**. ### What remains hard (the honest section)
- **Context**: **CLAUDE.md** quality varies widely across teams → strong impact on output.
- **Agentic security**: agents that **act** (not just suggest) → increased **blast radius**, security model needs rebuilding.
- **Roles**: juniors→seniors if AI absorbs entry-level work? role of the designer/PM? **execution unit** scrum team → experiments with **1 or 3 people**. ### To use in engagements / presentations
- Serves as a **proof point** for the *Token & Outcome* deck (the "voice from the field" / "frugal car" slide): a hyperscaler **removes** the limits and **gains** in quality.
- Convergence triangle: **Salesforce (operational proof)** + **Gupta (economic framework)** + **Greenwald (pricing model)** = the same message: **manage the outcome, not the token**.

## RésuméDe400mots

Srini Tallapragada (President & Chief Engineering Officer at Salesforce) published a *progress report* on May 27, 2026: after crossing 90% AI adoption, Salesforce Engineering moved from "copilot" usage to a **genuinely agentic SDLC**, where autonomous tools write code, review PRs, generate tests, update documentation, and manage deployments.

The inflection point: **org-wide standardization on Claude Code** and, above all, **removing all token limits**. The doctrine: the token limit is *friction* to eliminate, not a budget safeguard. The results (April 2026 vs 2025): **+50.8%** work items per developer, **+79%** PRs merged, and an **Effective Output score** (an ML measure of the code's **real value**, not volume) **+151.3%**.

Proof by example: a migration of **33 API endpoints** to a cloud-native architecture, estimated at **231 person-days**, completed in **13 days — 18× faster**. The method: a *rule-based* framework built in Claude (markdown + reference implementations) whose rule set grows with every PR feedback, **autonomous LLM loops (build, fix, validate)** with no manual intervention, parallelized across isolated environments. Outcome: **5 PRs**, the largest delivering **21 endpoints with 100% coverage**.

Against the idea of a speed/quality tradeoff, the **Engineering 360** platform shows **incidents dropping by 5%** despite the rise in PRs: *"quality doesn't suffer from speed. It benefits from it"* — thanks to security guardrails and quality standards **structurally embedded** in the workflow (Trust as the #1 value).

Beyond the numbers, Salesforce is **overhauling the SDLC**: which processes to eliminate, which handoffs to remove, what human work can an agent own? A **new craft** emerges: **Claude Code skills** become a shared engineering artifact; the **AI Expert Suite** and **Salesforce Foundation Plugins** institutionalize a skills library (more accuracy, less unnecessary cost); **subagents and agent teams** parallelize workstreams — the engineer *describes the outcome*, the agents find the steps.

The author acknowledges what remains hard: **context management** (variable CLAUDE.md file quality), **agentic security** (agents that act → increased blast radius), and **evolving roles** (becoming senior, the role of designer/PM, the execution unit shrinking to 1 or 3 people). Conclusion: the transformation *"changed what was economically possible"*; the ambition is to build *"the most automated, agentic SDLC in the industry"*. A major empirical piece that validates, from the operator's side, the shift from token to outcome.

## GrapheDeConnaissance

- Srinivas Tallapragada —dirige→ Salesforce (ORGANISATION, 0.97)
- Salesforce —utilise→ Claude Code (TECHNOLOGIE, 0.97)
- Salesforce —affirme_que→ « we removed all token limits » (CITATION, 0.98)
- suppression des token limits —améliore→ output et qualité (CONCEPT, 0.9)
- Effective Output score —mesure→ valeur réelle du code livré (CONCEPT, 0.92)
- Effective Output score —mesure→ +151,3% en glissement annuel (MESURE, 0.95)
- workflow agentique —permet→ migration de 33 endpoints en 13 jours (EVENEMENT, 0.96)
- migration agentique —mesure→ 18× plus rapide que l'approche manuelle (MESURE, 0.94)
- Engineering 360 —mesure→ baisse des incidents de 5% (MESURE, 0.92)
- Srinivas Tallapragada —affirme_que→ la qualité bénéficie de la vitesse (AFFIRMATION, 0.9)
- Claude Code skills —est_instance_de→ artefact d'ingénierie réutilisable (CONCEPT, 0.88)
- Salesforce Foundation Plugins —réduit→ coût inutile (CONCEPT, 0.85)
- subagents / agent teams —permet→ parallélisation des workstreams (CONCEPT, 0.88)
- Salesforce —affirme_que→ la qualité des fichiers CLAUDE.md pèse fortement sur la qualité de l'output agentique (AFFIRMATION, 0.86)
- Salesforce —affirme_que→ la sécurité agentique exige un modèle de sécurité fondamentalement différent (AFFIRMATION, 0.88)
- transformation agentique —permet→ ce qui n'était pas économiquement possible auparavant (CONCEPT, 0.85)

---
Canonical: https://www.thekb.eu/en/fiches/salesforce-tallapragada-how-engineering-became-agentic-2026-05-27/