From Projects to Platforms: Why Connected Intelligence Will Redefine Differentiated Services
The Services Model Is Being Rewritten
For three decades, data and analytics services were sold on effort: skilled people, billed by the hour, building bespoke solutions. AI is dismantling that model faster than most firms realise.
When agents can draft requirements, generate test cases and write runbooks, effort alone no longer commands a premium. Clients will pay for outcomes delivered faster, with less risk, and with intelligence that keeps getting better.
The pressure is visible across the industry. Fast Company has spent much of 2026 chronicling how agents are reordering enterprise software itself. In one interview, ServiceNow CEO Bill McDermott argues the new moat may be workflow depth rather than interface stickiness. Services firms face the same reckoning. My conviction is that Connected Intelligence will decide which services and offerings stand apart over the next five years.
What’s Changing
Pilots Are Giving Way to Production
In a February Fast Company column, DeepL founder Jarek Kutylowski cited McKinsey data showing that 62 percent of companies are experimenting with agents, yet only 10 percent are scaling them across a single function. The gap between leaders and laggards is now one of execution, not ambition.
Client Patience Has Compressed
A sponsor who once accepted 12 to 15 months to a first AI output now expects one within a quarter.
Platforms Have Converged, but None Will Rule Alone
Most enterprises have invested in Databricks, Snowflake, AWS or Azure. The platform is no longer the decision; what runs on it is. Fast Company’s report on Box CEO Aaron Levie’s conversations with more than 20 enterprise AI leaders makes the same point: enterprises run many AI systems, and interoperability is shaping the stack.
Delivery Itself Is Being Automated
Repeatable work such as BRDs, test cases and runbooks absorbs about 40 percent of engineering time on a typical programme. Agents can now take much of it on.
Boards Want Proof
AI budgets face the same scrutiny as any capital programme, and leaders expect to see which initiatives earn their keep.
What Doesn’t Work
Adding People to the Same Problem
Firms that answer AI by staffing larger bespoke builds will see margins and relevance shrink together.
Starting from Zero Every Time
When each engagement rebuilds connectors, data models and governance, the client pays repeatedly for the same foundations, and nothing compounds. The result is sprawl. In a September Fast Company column, OneTrust’s Blake Brannon cites Gartner’s forecast that the average Fortune 500 company will run more than 150,000 AI agents by 2028. Built project by project, that is not a capability. It is a liability.
Governing After the Fact
Brannon also notes Gartner’s finding that only 13 percent of organisations believe they have the right AI governance in place. Review processes designed for slow-moving software cannot keep pace with agents that make thousands of decisions before a committee meets.
Locking Clients In
Architectures tied to one model provider look efficient until the market moves. Then a switch costs three to six months and up to $1 million in re-architecture.
One Pitch for Every Client
A digital native with no data platform, a mid-way adopter on Databricks and an advanced user running twenty AI projects face different problems. Treating them alike wastes everyone’s time.
What Works
The firms that lead will build their services on connected, reusable intelligence. At Infocepts, our CoIN platform reflects five convictions.
Codify What You Learn
Every engagement contributes certified connectors, industry data models, prompts and agent patterns to a shared registry. The next client starts 40 percent of the way there.
Meet Clients Where They Are
Some need a full AI stack activated in weeks. Others need an intelligence layer on the platform they already own, or help governing a sprawling portfolio. The components are shared; each solution is assembled around the problem and the persona.
Govern In-Line, From Day One
Guardrails, explainability, drift monitoring and audit trails run with every deployment, not after it. That is how pilots survive security review and reach production, and how trust becomes a reason to buy rather than a reason to delay.
Stay Portable
Solutions run on the platform of the client’s choice, and a routing layer keeps them independent of any single model provider. A provider change becomes a configuration update measured in days.
Put Agents to Work on Delivery
When agents handle repeatable tasks, senior engineers focus on the problems that differentiate the client.
The economics shift too. Year-two projects on a mature registry cost 40 to 50 percent less than year-one equivalents. Value compounds for the client and the provider alike.
What to Ask Next
Whether you buy AI services or provide them, four questions deserve board-level attention.
1. Are Your Partners Selling Effort or Accumulated Capability?
Ask what they reused on your last project, and what that project left behind for the next.
2. Is Your AI Portfolio a Platform or a Collection?
If you cannot see cost and value per use case, it is probably a collection.
3. Can Your Governance Move at Agent Speed?
If oversight still depends on human review after deployment, scale will outrun control.
4. Are You Preserving Choice?
Your platform, your models and your data should remain yours to change.
The next generation of differentiated services will not be defined by the size of the bench. It will be defined by how much intelligence each engagement inherits, and how much it leaves behind.
Kumar Amitesh is President of Infocepts.
Website: www.infocepts.ai