Port CTO says a faster coding agent will not fix the PR queue
In an October 10th CTech opinion piece, Yonatan Boguslavsky argues that teams need shared context, workflow controls and clear human handoffs to coordinate agents.
By RuntimeWire Staff · Published
Primary source: Ctech
Why it matters
Boguslavsky is making the case that agent adoption depends on organizational context and workflow controls as much as coding ability. That positions Port to sell coordination across changing agents. Its test is whether the controls remove handoffs without adding process.
Port co-founder and CTO Yonatan Boguslavsky argues that engineering teams trying to scale AI agents should first fix the work that stalls between them: missing service ownership, context trapped in separate tools and reviews waiting for a person to decide what happens next. In an October 10th opinion article for CTech, Boguslavsky calls a system that coordinates those handoffs a software factory and says teams should start with their actual bottleneck.
His examples are familiar operational snags. An alert fires, and an engineer copies the error into an agent prompt. A ticket says "fix login bug," leaving someone to work out which repository it concerns. An agent opens a pull request, then it waits because the service owner is unclear. In each case, the agent can produce work; the surrounding engineering system still depends on people to supply context or move that work along.
The founder's argument comes from platform engineering
Boguslavsky has worked on this category of problem before. Before founding Port, he led an infrastructure group in Israel's Unit 8200 that built an internal developer portal used by thousands of developers, according to Team8's profile. The portal helped developers access private-cloud resources and DevOps services. He also worked with Hadoop, Kubernetes, cloud infrastructure, containers, microservices and automation, Team8 says.
His argument concerns the organizational plumbing that lets engineering teams use shared services: who owns a system, what an automated process may touch, which rules it must follow and when a human must take over. The same problems that made internal developer portals useful become more consequential when software agents can act across repositories and production workflows.
Boguslavsky's proposed factory combines rule-driven steps with decisions an agent can make as work evolves. A team could define deployment gates around service criticality, incident status and release timing, then send a failed condition to a human. An agent might discover a requirement missing from a feature specification, pause the work and ask a product manager to approve a change. For a production regression, a monitoring rule could trigger a rollback while an agent investigates and prepares a fix for review.
These are examples in an opinion piece, not reported customer outcomes. A workflow can be drawn as a neat chain of agents, checks and approvals, while real teams still have to decide what context to share, who is accountable and how much authority to grant. Boguslavsky frames that last decision as a balance. Too many escalations leave engineers managing agents; too little oversight can turn review into a rubber stamp.
Port is selling the layer around the agents
Port's commercial position follows the same argument. Port describes its platform as a way to build, govern and operate the software development lifecycle. Its current product materials include a software catalog, shared context, workflow orchestration, agent management and governance. Port also describes integrations that bring external agents into a catalog and let teams invoke them through workflows, rather than requiring each team to wire up a separate process.
That puts Port in a different position from a vendor whose central product is one coding agent. Port is pitching the connective layer around agents that a company already uses: the catalog of services and owners, the permissions and policies that constrain actions, and the workflow that routes work between automation and people. The bet is that those organizational records and controls stay useful as individual agents change.
Boguslavsky's advice is to find where work waits longest. If verification and code-owner review are the slow steps, a faster code generator will not clear the queue; ranking reviews or grouping related pull requests may help more. As teams automate individual steps, he says, they should reuse shared context, orchestration and controls so each fix contributes to a system that works across the organization.
The strategy is also visible in Port's financing history. In December 2025, Port announced a $100 million Series C led by General Atlantic, with Accel, Bessemer Venture Partners and Team8 participating. TechCrunch reported that the round valued Port at $800 million and brought its disclosed funding total to $158 million. Those figures describe the earlier financing, not a new deal connected to Boguslavsky's CTech article.
The funding gives Port room to pursue a category beyond developer portals: managing and coordinating agent-driven work across software teams. Port's challenge is to make that coordination reliable without adding a layer of process that slows customers down. Boguslavsky's essay recommends measuring handoffs, identifying waits and automating the specific point where a person is still pushing work from one step to the next.