Agentic Engineering

AI accelerates,
people steer

Agentic Engineering is how we use AI in every project at Foreside. Not as a magic wand, but as a powerful assistant that takes the routine work, so our engineers can focus on the things that actually move the needle.

Schedule a meetingTry the Kompas
03, Agentic Engineering

AI accelerates our work
Our people make
the difference

Time-to-market
Measure
Per project, before and after
Production incidents
Track
After every go-live
Maintenance cost
Monitor
Per team, per quarter
How we use AI in practice
01

Faster scaffolding

AI generates boilerplate, test scaffolds and configuration. Our engineers review, refine and own the result.

02

Safer code

AI surfaces edge cases and vulnerabilities at PR time. Humans decide what ships.

03

Living documentation

Docs and decision records are generated alongside code, not as an afterthought.

Architecture

A knowledge base
on your own sources

A knowledge base answers questions from your own documents, case files and conversations. You only see what you are allowed to see, and every answer says where it came from.

How a knowledge base worksOn the left the knowledge base is filled from your own sources, at the bottom someone asks a question from their workplace, the search step fetches the passages that match that question and fall within their permissions, and the answer comes back with its source.FILLING THE KNOWLEDGE BASEUSING THE KNOWLEDGE BASEQUESTIONPASSAGESANSWERYour sourcesmanuals, contracts,case files and meetingsCollect and processtext extracted, cut into chunks,with permissions and originThe knowledge basesearchable by meaning and bykeyword, always up to dateYour workplacechat, portal, Teams or your appthat is where you askSearchthe passages that match thequestion, within your rightsAnswer with sourcewith the passage attached, soyou can check itPermissions, freshness and loggingyou only see what you may see, a changed document goes through the mill again, and every question can be traced backThe heart of the solutionBack to where you started

Swipe sideways to see the whole picture.

Architecture

An agentic
workflow in production

A workflow carries out work instead of answering questions. The agent runs a round of looking up, carrying out and checking, and keeps going until the result passes its own test. It stays within your boundaries, and for budget, risk or doubt you decide.

How an agentic workflow worksWork comes in and gets routed, the agent makes a plan and then runs a round of looking up, carrying out and checking, until it holds. It draws on tools and memory and stays within fixed boundaries. Then a person decides, and only then does the result land in your own systems.COMING INTHE AGENT AT WORKGOING OUTROUTEDOES ITANOTHER ROUNDPROPOSALAPPROVEDThe triggera mail, a ticket, an order, aform or a fixed momentTriagewhat is this about, how urgentis it, and which path fitsThe plancut into steps, with a budgetin time and in costLook upreads your systems and theknowledge base, never moreCarry outdrafts it, fills the fields,writes the codeChecktests it, holds it against therules, lets a second model lookYou decideyou see exactly what willhappen before it happensCarried outin your own systems, with away back when it goes wrongREAD AND WRITEToolsthe knowledge base, your ownAPIs, documents and mailGETS BETTERMemorywhat was decided and whatworked, so the next run is fasterBoundaries, budget and stop buttonWhat the agent may and may not do is fixed, there is a ceiling on time and cost, every step can be traced back, and you can always stop a workflow.In practice: preparing quote requests, triaging tickets, checking invoices, handling onboarding and preparing monthly reports.Where the choices are madeWhat the agent draws on

Swipe sideways to see the whole picture.

How we work

Several agents,
each with one task

One agent doing everything delivers average code. So we cut the work up: every role gets its own agent and the model that fits the task, and whoever builds never reviews their own work. Only when build, lint and tests are green and the review is clean does anything go live.

How good code comes outA chain of agents, each with its own task and its own model: explore, plan, build, test, review and only then live. When a gate fails the work goes back to building, and whoever builds never reviews their own work.THE AGENTS AT WORKRED? BACKFINDINGSREADS ONLYExplorewhat is already there andwhich pattern fitsHEAVIEST MODELPlanthe story and the technicalspec, before any codeTHREE AGENTSBuilddatabase, backend andfrontend, one agent eachFROM OUTSIDETestbuild, lint and tests green,and the page really usedANOTHER AGENTReviewreads only, checks theconventions and criteriaGREEN ONLYLiveonly from main, only wheneverything is greenYou decide at three momentsAfter the story, after the spec and after the final report. In between the chain runs without you.Every role gets the model that fits the task, and nobody marks their own homework.Where it stands or fallsNot green? back into the chain

Swipe sideways to see the whole picture.

Want to see Agentic Engineering in action?

Try the Foreside Kompas, or schedule a meeting and we will show you a project.

Schedule a meetingTry the Kompas