By Dwayne McDaniel, Senior Developer Advocate at GitGuardian
In 2027, we will see a change in the enterprise about where software comes from, who creates it, and how security controls are drawn and managed
AI agents are giving employees the ability to build software, connect systems, and automate work outside the development processes enterprises spent the last twenty years securing. They are also gaining enough access and autonomy to force a deeper conversation about identity, authorization, and intent.
Security teams will still protect traditional applications and infrastructure. They will also inherit a second software estate that grows faster and behaves very differently.
Prediction 1: Citizen coders will create more of the software their departments rely on
By the end of 2027, a large share of the small applications, automations, and workflows employees use every day will come from citizen coders.
Much of this software will bypass traditional development processes. There may be no shared repository, CI/CD pipeline, or formal security review. It may run on localhost or inside a shared agent platform while still connecting to production SaaS, company data, and internal APIs.
This software will be created increasingly by employees with no formal developer or security training who have added skills, instructions, and integrations when they see value. Analysts will teach agents to move information between systems. Finance teams will build workflows that combine internal data with third-party services.
Centralized, shared AI platforms will accelerate this. Employees in each department, far more separate than traditional developer teams, will be asked, and sometimes mandated, to leverage systems that have enterprise permissions. Local agents will extend the same model onto employee machines, using existing credentials and authenticated sessions to reach real data and services.
We are creating a software estate outside the places AppSec and DevOps teams traditionally monitor and manage. Security teams are increasingly going to be asked to solve for this reality where repositories and pipelines will remain critical, but they will represent a smaller share of where software is created and operated.
The definition of a developer will expand in 2027 to anyone who can describe useful work to a machine.
Prediction 2: AI will finally force enterprises to connect identity with intent
Enterprise identity systems have traditionally answered who or what has access.
AI agents will force security and DevOps teams to also ask what that identity is trying to accomplish right now.
Security teams are well aware that agents can start with a legitimate task and discover their own path to complete it. The team members using coding agents are not fully aware of the risks that come when their AI-driven systems can read files, call APIs, or interact with other services in sequences no one explicitly designed.
In response, we will see enterprises accelerate adoption of workload identity, short-lived access, isolation, and tighter network segmentation around agent workloads. Technical advancements and hardening of standards for multi-system environment workload identity will make it far easier to adopt these patterns, even for teams that could not have approached this subject even a few years ago.
Across SaaS platforms, internal systems, and public networks, agents will be able to reach whatever their available access allows, and that will continually expand if we make no change to how we deal with access. Sensitive actions must increasingly require authorization based on what an agent is doing now and what it did moments before. Context will be king for how we authorize.
Identity, authorization, and intent will start operating as one security model far beyond the most mature enterprises in 2027.
Prediction 3: Employee endpoints become much more valuable targets
As millions of employee laptops start looking more like developer machines, endpoint attacks will change with them.
Traditionally, infostealers have been designed to exfiltrate passwords, browser sessions, and cryptocurrency wallets. Moving ahead, as more citizen developers create code locally, their endpoints will also contain more API keys, agent configurations, and authenticated connections to business systems. Attackers know this.
We will see an expansion of what infostealers look for, beyond the ability to log into the human path. Infostealers will increasingly target the files, configurations, and credentials that let agents operate. Stealing access from one employee machine opens paths into several systems at once.
Organizations will increasingly treat local AI environments as production-adjacent infrastructure whenever they can reach production data or services.
Prediction 4: Security controls will move from reviewing software to governing actions
Up til now, enterprise software security focused on source code, dependencies, and deployment pipelines. And those still need defending. But moving ahead, AI agents will increasingly act before those artifacts reach a review point. Agents can generate code, call APIs, and execute commands immediately, using previously granted permissions that might not have been for the task.
Security controls will shift closer to the action itself. We are already seeing this happen with agentic AI hooks and access policies enforced through sandboxing. Organizations will increasingly add deterministic checks before sensitive commands, API calls, and credential use. This will be the common approach by the end of 2027, but we are likely to see a lot more nightmare news stories along the way.
Prediction 5: It will be common for companies to pause and harden in the face of feature overload
2027 will see more deliberate pauses in feature development as companies realize they have shipped faster than they can safely sustain and at a rate no customer could possibly keep up with. New features will see diminished returns as more users will demand stability above all.
AI has meant faster time to features, integrations, and new service offerings. Besides exhausting users who are seeing this increase from all vendors, every addition creates more code to maintain, permissions to govern, and infrastructure to operate. Feature velocity eventually creates feature debt while making consumers nervous that their favorite features will change in dramatic ways.
We have already seen this in play as Epic recently paused most product development for several weeks to focus on security issues in its platforms. More companies will make similar choices. Engineering organizations will dedicate larger blocks of time to reliability, security debt, architecture simplification, and removing expensive features.
Ironically, AI, while driving this issue, also lowers the cost of hardening, and pausing to catch up on the PR queue to improve the app or service will become a common part of the product lifecycle.
2027 will push security beyond the pipeline
Professional, dedicated, traditional engineering teams will keep building critical applications while “citizen coders” will create workflows, agents, and small applications outside of traditional IT.
Security teams will continue to respond to this evolution. Endpoints will increasingly become part of the secured development environment. Identity will become inseparable from intent when it comes to authorization. Policy decisions and enforcement will continually move closer to individual actions.
In 2027, anywhere an employee can give an AI agent a goal can become a development environment. The organizations that adapt fastest will stop assuming software, authority, and risk live in a few well-managed places.
