AI-First Culture in the Enterprise: What Ably Discovered

We are at the inflection point of AI adoption in enterprises. Enterprise AI spending reached $37 billion in 2025 — 3.2 times the 2024 value. At the same time, 70-85% of AI projects fail according to PMI and HBR.

Ably, a realtime infrastructure company processing 2 trillion operations per month, decided that to build credible AI products, they first needed to adopt AI internally. The result was not just tools — it was a cultural shift where AI usage became expected, not optional.

The Central Discovery

“The biggest gains come from how people think, not tools.” — Jamie Newcomb, Ably

This sentence summarizes everything. Companies that treat AI as a technology project fail. Companies that treat AI as cultural transformation thrive.

Ably’s Three Pillars

Ably structured their adoption across three fronts:

Internal Adoption: Integrating AI into workflows across all teams. Engineering uses Claude Code. Marketing analyzes calls with AI. Finance automated reconciliation.

Developer Experience: Making the platform more discoverable through AI-enhanced documentation and tooling.

AI Product Enhancement: Understanding customer use cases and building infrastructure informed by real needs.

The critical point: the infrastructure philosophy powering internal AI adoption also powers the AI Transport product Ably offers to developers.

MCP: The Invisible Infrastructure

The most significant internal development was an MCP connecting 15+ internal services through 140+ tools.

The Model Context Protocol is an open standard from Anthropic that allows AI systems to securely connect to tools, data, and enterprise systems.

Before MCP: “Help me write this customer email” — and you need to explain the company tone, infrastructure, guidelines…

With MCP: “Help me write this customer email” — the AI already knows all that. Context is connected.

Ably connected GitHub, Jira, Confluence, Slack, HubSpot, Gong, Metabase, PagerDuty, GSuite, and Jellyfish. A “tool registry” allows AI to discover only what it needs per task, solving context limits.

The Elephant in the Room: Security

MCP prioritized interoperability over security. This created significant risks.

Critical statistic: Deploying just 10 MCP plugins creates a 92% probability of exploitation. With 3 interconnected servers, risk already exceeds 50%.

OAuth 2.0 arrived only in March 2025, refined to OAuth 2.1 by June. Thousands of MCP servers deployed without authentication remain in production.

The “Shadow Agents” problem is real: just like “Shadow IT” in early cloud computing, enterprises now face unsupervised agents running on developer laptops accessing critical systems.

Results by Team

Engineering

The team uses Claude Code for agentic development with TDD guardrails. Implemented workflows include:

Discovery: “Where is X used?” — navigating complex codebases
CI/CD: Agent SDK integrated for PR reviews and issue fixes
Development: Initial scaffolding and code analysis
Documentation: Automated generation and maintenance

“A single human author owns every PR, regardless of AI contribution.” — Ably Policy

Anthropic data shows task complexity increased from 3.2 to 3.8 (1-5 scale) in 6 months. Feature implementation jumped from 14.3% to 36.9% of Claude Code usage.

Marketing and Sales

Marketing uses Gong call analysis for market research, automated lead validation across 6+ sources, and multi-stage ICP scoring with 8 criteria evaluation.

Sales implemented multi-signal lead routing, personalized email sequences based on ICP analysis, and expansion opportunity detection when customers approach usage thresholds.

Finance

Automated Stripe-to-Xero reconciliation eliminated thousands of monthly manual clicks. Direct spreadsheet creation through Claude using MCP data retrieval — no human intermediaries for routine tasks.

The Key: Enablement, Not Mandate

Instead of leading with AI, teams asked: “What repetitive processes exist?” — then explored AI solutions.

What Ably implemented:

Weekly drop-in sessions for questions, experiments, and collaborative problem-solving
Internal Slack channel documenting wins and experiments
Implicit mandate: Everyone should use AI, but framed as enablement
Focus on real problems, not technology for technology’s sake

“The cultural shift was from ‘Can AI help?’ to assuming it can, then identifying problems to solve independently.” — Jamie Newcomb

The Executive-Employee Gap

89% of C-suite believe they have an AI strategy. Only 57% of employees agree.

75% of C-suite think the rollout was successful. Only 45% of employees agree.

With training + leadership support, employee positivity toward AI jumps from 15% to 55%. The gap is not capability — it is enablement.

Why 70-85% of Transformations Fail

Most failures are not technological — they are people, processes, and politics.

70% fail when led by IT
95% of GenAI pilots fail (MIT)
46% of POCs scrapped before production
42% of companies abandoning AI initiatives (was 17% in 2024)

Failure patterns identified:

Delegation to IT: Leaders do not own the transformation
Lack of training: Tools without enablement
Isolated pilots: No scale strategy
Ignored resistance: 45% of employees resist or oppose

2-5x difference in expected value when AI is centrally orchestrated vs. distributed without coordination.

A4 Framework for Transformation

A1 - Assess: Diagnose the “organizational pathology” that kills initiatives. Map repetitive processes, identify resistance, evaluate data maturity.

A2 - Architect: Design a system where data flows freely across boundaries. Define governance, establish MCP/integrations, create security standards.

A3 - Activate: Deploy AI using “90-Day Sprints” from pilot to production. High-impact use cases, quick wins for momentum, fast feedback loops.

A4 - Amplify: Embed AI into culture as the default for decision-making. Scale company-wide, create AI Champions, measure and communicate ROI.

Proven ROI

$3.70 average ROI per $1 invested (top performers achieve $10.30)
74% achieve ROI in the first year
39% report doubled productivity
26-55% average productivity gains

High performers are 3x more likely to fundamentally redesign individual workflows. It is not about tools — it is about rethinking how work gets done.

Key Takeaways

Culture before tools: Mindset change is the differentiator
MCP is critical infrastructure: Connecting systems eliminates context friction, but security cannot be neglected
Enablement > Mandate: Training + leadership support increases positivity from 15% to 55%
Governance from day 1: 92% exploitation probability with 10 MCP plugins
Central orchestration is worth 2-5x: Distributed adoption without coordination loses significant value
Humans in control: Ownership and accountability remain human

At Victorino Group, we implement governed agentic AI for companies that cannot afford to fail. If you need to build an AI-First culture with strategy, security, and measurable results, let’s talk.